Do AI detectors work? Students face false cheating accusations
bloomberg.com
bloomberg.com
The only solution is to require face-to-face proctored exams and not allow students to use technology of any kind while taking the test. But any teacher doing this will end up with no students signing up for their class. The only solution I see is the Higher Learning Commission mandating this for all classes.
But even requiring in person proctored exams is not the full solution. Students are not used to doing the necessary work to learn. They are used to doing the necessary work to pass. And that work is increasingly cheating. It’s a clusterfuck. I have calculus students who don’t know how to work with fractions. If we did truly devise a system that prevents cheating we’ll see that a very high percentage of current college students are not ready to be truly college educated.
K-12 needs to be changed as well.
I was in high school when kids started getting cell phones with internet access and basically as soon as that happened it opened up rampant cheating even among the best of students. I can only imagine it being much worse nowadays than even 15 years ago when I was in high school.
Isn't it to either do that now, or to lose the signaling value of college degrees as indicating knowledge.
I'm not sure knowledge is what a college degree signals to prospective employers. The alternative hypothesis, which AFAIK has a fair bit of support, is that it signals a willingness to do whatever it takes to fulfill a set of on paper requirements imposed by an institution, by hook or by crook.
If actually learning is valuable to them, independent of whether it will actually help them with prospective employers, then yes. But I don't think we can assume that all students value that.
For professional engineers, at least, the degree is not what tells you whether they have learned the fundamentals. The license is.
Even for engineers in domains where there is no licensing (such as software, for example), I would expect them to have learned the fundamentals on the job, not in college. I think most employers expect the same; they don't view the college degree as a signal that the prospective employee knows the fundamentals, they expect them to learn that on the job. What the degree signals to the employer is that the prospective employee will comply with their corporate process.
While this is true, I’ll take the engineer who studied diligently with an aim to learn the material over the one who was just in it for the credential. I don’t agree with your implication that you will learn the fundamentals equally well on the job. It certainly happens, but from what I’ve seen it’s usually people who never had the opportunity to learn in school who are aware of their gaps and seek to fill them in. The nonchalant attitude towards theory tends to persist in the workplace.
The former engineer will learn on the job just as well. The latter one won't learn well in either environment.
You appear to believe that the crucial factor is the person's attitude towards learning, and I agree with that. I just don't agree that a person with that attitude towards learning is necessarily any better off going through college instead of getting a job and learning there. Particularly now, with so much good material available for free online, someone who wants to learn can do it without spending anything more than the cost of Internet access. So the opportunity to learn through college is even less beneficial for those who really want to learn than it was in the past. It might be enough of a benefit to justify college for some, but I think that number is much less than the number who actually go to college.
From the employer's perspective, if you think you can evaluate someone's attitude towards learning well enough, I don't see why their having a college degree would matter much one way or the other. If you think they have the right attitude towards learning, hire them! Whatever they don't know yet, they'll learn.
In Germany, all exams are like this. Homework assignments are either just a prerequisite for taking exam but the grade is solely from the exam, or you may get some small point bonus for assignments/projects.
> But any teacher doing this will end up with no students signing up for their class.
The main courses are mandatory in order to obtain the degree. You can't "not sign up" for linear algebra if it's in your curriculum. Fail 3 times and you're exmatriculated.
This is because universities are paid from tax money in Germany and most of Europe.
The US will continue down on the path you describe because it's in the interest of colleges to keep well-paying students around. It's a service. You buy a degree, you are a customer.
Same in the U.S. but you can sometimes find an online offering. If you don’t know what you are doing or don’t care then always take the online offering. Much easier to cheat.
My ex-girlfriend is German. She cheated on her exams to get her agricultural engineering degree at university. This was in the 80s.
The course might be mandatory but which professor you choose isn't. What if multiple professors teach it? Word gets around and everyone chooses the easy profs.
There were exceptions to this rule (in both directions), of course.
I guess there is some sort of incentives that rewards institutions taking the easy way out.
There is one Linear Algebra course. You have to pass it to get your degree. Typically, it's taught by the same prof for many years, but it might also rotate between different chairs and profs (but only one in each semester and the "design" and requirements of the course stays largely the same).
You don't have the choice to not take the class, you just have choice with which professor you would like to take it with. And often you would have to get lucky anyway, since that session may be filled so you'd have to take it with the "harder" teacher anyway.
For example with the popularity of computer science and STEM in general, at my school there were often 2-3 teachers teaching linear algebra in any given semester. And same for popular classes like calculus or introductory physics. Students would often lookup online which teacher was considered easier, but they still had to take the class.
Especially something like Linear Algebra can easily have class sizes of 800+ people at big universities. Yes there is typically exactly one lecture hall for that and you have 30+ exercise groups. But still only one faculty
But I'll say where I went to school, and I hear its even worse now since enrollment in STEM is way up, there were often multiple thousands of students every quarter wanting to take just one class, so they split it up because we simply didn't have lecture halls with enough seats. There would often be 3-4 classes each of 500+ students all full, and still students struggling to get in due to the maximum amount per course. Usually there was around two teachers splitting the sessions, and they also have their other more advanced courses and/or research.
So its probably just practicality in terms of their time and resources. This wasn't an issue with more advanced courses where there was usually only one teacher per semester offering the class.
Having the choice between different professors with supposedly different difficulties for what is supposed to be the same course seems absurd.
But I also mentioned that there are often thousands of students all trying to take one course. And the schools simply don't have the space to fit all of them in one session since I believe the rules are basically that it needs to be held in a lecture hall big enough to fit every enrolled student, and teachers don't have the time to teach 4 different sessions by themselves on top of their other duties. Maybe class sizes are just smaller elsewhere, but where I went to school it was not unheard of to have multiple thousands of students needing to take one class that was required for practically every STEM major in a given semester.
There's a line of research that shows that high quality one-on-one instruction gets you up to 2stdev gains in learning performance.
If you can afford to increase the professor to student ratio and make them available for office hours, you probably do see increases in performance. Is it due to better motivation? Seeing an academic up close? Actually better explanations you get from an expert in the subject? Hard to say.
So schools offer multiple sections of the same class to share the workload. E.g. in recent years Computer Science 101 - often the most popular class on campus.
Remember, in European countries students are admitted to study a specific subject at university, rather than being admitted to the university as a whole and expected to choose a major later on.
So there are multiple courses going on, with a lot of intersection between the topics covered. There's maths for computer scientists (heavy on the discrete maths), maths for engineers (heavy on the integrals and matrices), maths for social scientists (heavy on the statistics), and so on.
So both American and European universities split their year 1 maths courses so they can get a few thousand first-year undergraduates through the largest 300-500 seat lecture theatres. But in Europe it's a split by subject, rather than by choose-your-instructor.
This is true in the US as well. You can change your major, but you are admitted into a College in the University. Moving to another College is not guaranteed if you later change your mind.
But that's just my personal experience. I don't know if it's different at other large universities.
So roughly 6x the amount of students.
wat
You point to a true failure in incentives. And yet, the US has the highest density of renowned universities.
I didn't realize that so many community colleges and state universities were basically online diploma mills.
The renown has to do with a lot more than demonstrated ability of graduates.
And at least when I was in college it was the same with respect to classes, you can't take the same class more than 3 times. Additionally if a course is required you either take it or make the case for an equivalent class.
Almost every university in the US takes federal money and relies on federal loan guarantees to keep the high revenues pumping through. In exchange, the schools are subject to requirements by the government and they impose many. I think the bigger issue is the size and scope of higher ed here and if it's actually a good idea to to tell every school how to run their exams (and enforce it).
Sure, but they don't set the rules; sure, they do much of the education, but much of the demand for them comess from bachelor’s-degree bound students, so the course selection is set by what bachelor’s degree granting institutions accept.
The Higher Learning Commision is a farce. It’s purpose is for sinecures for its employees.
Very strongly depends on the school and major; there are both narrow-path degrees with lots of mandatory courses and wide-path degrees with very few specifically mandatory courses (instead having several n of m requirements) other than lower-division general education requirements.
...which is what we have today, where the most lucrative industries for people with good math skills are finance (= cheating dumb people out of their retirement), advertising (= cheating dumb people out of their consumer dollars), and data-driven propaganda (= cheating dumb people out of their votes).
/dystopia
I like the phrasing you used.
Advertising absolutely works on you regardless of how smart or educated you are.
How it has to work to do that can change, but the idea that advertising only impacts dumb people is pernicious as shit.
This system truly forced students to grind the hell out of science
An argument perhaps that there should not be hundreds in a room.
Long term I see education going this route, rather than preventing students from using AI tools, update course curriculum so that AI tools don't give such an advantage.
The real issue as I see it is that no one wants to face the reality that far too many incapable, incurious people are going to college. So I pretend to give real tests and pretend to give real grades and students feel good about themselves and my classes fill.
Fundamentally, kids that are just trying to pass a class don't see the value in learning and it seems that the contributions towards the "pointless" school work are parts teacher attitudes, parts curriculum design, parts real-life applicability to the student's interests, parts framing.
We've been using tests and such for far too long as a proxy for competence, rather than developing the competencies in such a way that engages the kids.
I think we need to look at reframing fundamental parts of how education is structured. I don't think there needs to be drastic changes, just some small things that allow the education and curriculum to become more engaging.
I think the vast majority of people who say and think this haven’t taught in the classroom much.
It is 100% societal. It is because society is focused on "get degree, get job, get money". It is because Western societies have gotten so damn competitive that if you don't succeed at any of the above, there is a non-trivial chance you won't even be able to afford a house to live in.
In America, I'll admit that No child left behind made it a lot worse, with tests left and right, which gives students the wrong impression of what learning is about.
Every class should be about critical thinking. Every single class. Multiple choice tests are a societal cancer and should be limited to a tiny fraction of tests given.
The point of school is to learn how to learn. That is it. What facts are taught are almost irrelevant. The point is to learn HOW to learn. Be that researching the history of fabric dyes in Ancient Egypt or making a scale drawing of one's house.
The "WHAT" IS NOT IMPORTANT.
The HOW is important.
How to write an essay, the topic doesn't matter.
How to learn about the culture of a country.
How to learn a new field of mathematics.
How to learn a new type of art.
How to give a presentation.
How to learn a hard science.
Yes the basics of physics and chemistry and such need to be taught. But the things that are learned should be inline with teaching the all important skill of how to learn.
Students want to be engaged in their coursework, but the universities aren't there to encourage or support it.
That seems more of an indictment of your profession than anything to do with the students.
Students largely just want to pass. They mostly don't care that they don't know anything.
If the school has a tuition, then there's even more of a conflict of interest. I've had parents/admins imply that we might be losing a student due to poor grades.
Profs who were hard but fair never had a problem filling up their classrooms with students who self selected for wanting to learn.
The hard but not fair ones were just assholes IMHO.
The easy ones also had their classes filled up.
My community college had two history profs, one had all essay questions, one had multiple choice. The essay question prof was considered "hard", but so long as your essay justified your position and was well reasoned, you got full credit for the answer.
I hated the multiple choice prof. He gave the entire class his test bank every quarter and you just had to memory hundreds of questions and he'd pick 50 for the test. IMHO it took more time studying because I had to read the book and then memorize a bunch of pointless answers, vs reading the book and understanding what was going on, which I can typically do in the first pass.
The first was a second-year Physics teacher at my community college. He said that at least 50% of the class should fail. Not 50% will fail, but that 50% should fail, and if less than 50% fails, the class was too easy. He demanded that we memorize 30+ equations that could appear on the final and did not allow a notecard of equations.
The other was my Algorithms teacher in my third year, who basically wasn't teaching what we really needed to know, as far as I was concerned. I was expecting to learn breadth/depth-first searching, binary search, sorting algorithms, maybe even path-finding like A* and Dijkstra's, etc. Instead, the entire course was about algebra dealing with Big-O, Big-Theta, and Big-Omega, with only a slight nod to how they related to code.
Then we need to teach them. You are doing the right thing for being a "hard" teacher, and it doesn't prevent you from also being known as a caring one.
From experience, acknowledging the students' difficulties with it and emphasising that it is because they were not taught how to think (as opposed to some innate inability to do maths) can go a long way.
But in the (hypothetical) limit where AI tools outperform all humans, what does this updated test look like? Are we even testing the humans at that point?
This is because 100-200 level math courses are not about teaching anything, but about filtering out students who can't do the work. Once you get past that level students have already formed bad habits and so still only do what it takes to pass. I don't know how to fix it, I don't know if it CAN be fixed.
This is 100% incorrect.
Admittedly, they are about teaching things. For example, teaching Laplace transforms to mechanical engineers. It certainly isn't true to say the "courses are not about teaching anything".
But if 20% of the class should decide to change majors to business? Well, there's been some filtering out of students too.
The same for job interviews. I did a lot of technical interviews in the past as interviewer (hundreds) for Software Engineer positions (and still help companies to hire sometimes, as independent interviewer).
There is insane amount of cheating. I'd say at least 30% in normal companies are cheaters, and 50% and more in FAANG. I can prove it, in private groups, and forums people share tech assignments. And very large number of these people use some kind of assistance while interviewing.
It's interesting to see how sometimes questions that are intentionally sophisticated are getting solved in a few minutes the best way they can be solved. I see this over and over.
I interview a lot of people and I rarely see anything I'd describe as cheating. Maybe my company is not famous enough to be worth cheating at.
In university I can recall a computer graphics course where literally everyone got 100+% on problem sets (there were bonus questions of course) and the median score on the midterm was below 50%. Leading up to the exam I remember the prof leading an exam prep session, opening the floor to questions, and getting a sincere request from one of the students to please go over the whole concept of "matrices" again.
This was a 400 level course, BTW. At one of the highest-rated universities in Canada. (I was taking it as an elective from a different program from the default, so I can't speak to the precise prerequisites to get there.)
This was over 20 years ago, BTW. I'm sure it's only gotten somehow even worse.
The midterm and final exam were in-person in bluebooks, and they were 60% of your grade. If you were just copying the problem sets, you would fail the exams and likely the class.
They ended up changing the assignment to where you could just find an implementation of a FFT online and write about it or something.
That's not even getting into the students who copy-pasted Wikipedia straight into their papers in that same class.
As a student, the only thing the next institution will see is GPA, school, major. Roughly in that order. If the cost of not getting an A is exclusion from future opportunities- then students will reject exclusion by taking easier classes or cheating.
As someone who studied physics and came out with a 2.7 GPA due to studying what I wanted (the hard classes) and not cheating (as I did what I wanted) - I can say that there are consequences to this approach.
In my opinion, the solution is to reduce the reliance on assessments which are prone to cheating or which in the real world would be done by computer.
I know an equitable grading champion at an LAUSD school, I'll see if I can get material to share. EDIT: I just received [0][1][2][3].
[0] (5 page pdf) https://drive.google.com/file/d/1YO7SQEwisAbHHi6mfgj7XU9FcSB...
[1] (4m30s video) https://drive.google.com/file/d/10eWor4uhSxR8ZITA1w3kzqhTOX0...
[2] (audio interview) https://www.bamradionetwork.com/track/fair-grades-dropping-g...
[3] (article) https://ascd.org/el/articles/taking-the-stress-out-of-gradin...
The nice thing about an all pass/fail system is you can formalize the 'new' way grades are actually done in which A means meets expectations and anything less means did not. Making pass mean A/B takes a lot stress off students and C/D is already failing for practical purposes as often you can't continue with less than a B.
If it's a job, the order will be school, school, major, everything else on the résumé, grades maybe.
That's not the cost of not getting an A, it's the cost of appearing to underperform compared to too many of your peers. Which is directly tied to how many of them cheat. If not enough cheaters got an A then the cost would no longer be tied to not getting an A, it would be tied to whatever metric they appeared to outperform you on.
Really focusing on stretching yourself necessarily means lower grades. Why is that penalized? TBH, in software engineering a lot of people with lower grades tutor the ones with 4.0 averages. The skillsets required to code and the skillsets required to get a good grade on a test are different.
Maybe worse was seeing the undergrads who passed on research opportunities out of fear it would distract them from keeping a high GPA.
The follies of youth and all that :)
> Really focusing on stretching yourself necessarily means lower grades.
It's overly dramatic, but I think gets the point across in an easy to understand way. It's exactly why you see the lower grade ones tutor the high grade ones (this even happened in my undergrad and I did physics[0]).
It's because learning happens when struggling. It happens at the edge. This is also a big reason some learn a lot faster than others or even why someone will say they don't understand but understand more than someone who says they do (and who believes it). Because expertise isn't about the high level general ideas, it's about all the little nitty gritty details, the subtle things that dramatically change things. But a big concern I have is that this is a skill to learn in of itself. I think it's not difficult to recognize when this skill is learned (at least if you have) but it's not something that'll be learned if we focus to much on scores. After all, they're just a proxy. Even the institutional prestige is a proxy (and I have an argument why it no longer matters though it did decades ago).
I do wonder if this is in part cause for the rise in enshitification. Similarly if this is why so many are bad at recognizing issues in LLMs and ML models. I'm sure it is but not sure how much this contributes or if it's purely a confounding variable.
[0] when I signed up to be a tutor at my university I got signed off my the toughest math professor. When I took the signature to the department the admin wasn't sure if I was trying to trick her because she immediately called the professor to confirm the signature. Then told me I could tutor whatever I wanted because I was one of two people he had ever signed off on. Admittedly, I'm sure a lot of that was because people were afraid of him (he wasn't mean, but he wouldn't let you be anything less than the best he thought you could be)
My personal anecdote is a professor who was incomprehensible (complete failure on the teaching side; required class), but meticulous in pointing out all the important equations. I came to the proctored exam with the allowed one sheet of paper with all the "important equations", and solved the problems "expert system style", applying equations in whatever way appeared to solve the puzzles. And got a perfect grade while the rest of the class failed spectacularly. This was all according to the rules and I got the grade I wanted but complete wasted time as to learning.
Essentially when a scarcity increases, there will always be an imperfect heuristic of selection.
I guess this is more of a public policy area but it seems reasonable that anyone working full time should have access to economic security. Essentially cheating on university is the first symptom of lifetime of vastly unequal access to economic security.
Some organizations still hire software engineers just based on resume and a nontechnical interview. This can easily be a disaster! You need to do a real assessment during the interview of how well software engineers can code.
-> GPA can be gamed, as laid out.
-> Take Home assessments can mostly be gamed, I want to assess how you think, now which tools you use.
-> Personality tests favor the outgoing/extroverts
-> On-location tests/leet code are a crapshoot.
What should be best practice here? Ideally something that controls for first-time interviewer jitters.
The other thing you have to do is that you have to be willing to fire the people who are underperforming. It's just a natural consequence of the interview process being imperfect.
I think doing very basic, simple things is a better test. I also like asking candidates to explain something that they already understand well. And for code, I like reading it and also getting an explanation about why things were done the way they were. Code reading also seems to be an important skill that is probably more important than writing.
They aren't 'fair' in avoiding every false negative, but they at least tell me that the passing candidates know and can do something.
If I ask someone who claims to know Python or Java whether or not you can have a memory leak in them, and their answer is 'no' or 'Maybe, but I don't know how', I get a pretty good idea of whether or not they know anything about this topic.
If you can't do fizzbuzz, you probably aren't a good fit for a SWE position either, you should be aiming for something more director-level. Given how much people struggle with coding, I sometimes feel like I may as well ditch my regular question, and just ask them to write that.
When I was at Google, most of these were filtered out before I got to see them, but for a while I was doing iOS development interviews and a lot of the candidates applying to Google clearly didn't know anything.
Then I started interviewing, and... yeah. I get it now. It really is that simple, should take a competent coder a few minutes, and 80% of people interviewing will take 45 minutes to muddle their way through it.
Onsite interviews with a range of approaches seem to be the best I've found over the years. As much as it pains me, things like fizzbuzz are still useful, because people still lie about their ability to program in languages. If you claim to know C very well and can't knock that out in 5 minutes, and it takes you 45 minutes of prompting, well, you don't know C usefully.
I've seen good results with having a pre-done sort of template program that's missing functionality, and the person completes it out based on comments (for remote interviews), and you can generally tell by watching them type how familiar with the space they are. Again, perfection isn't the goal, but if someone claims to know C very well and is trying to make Javascript syntax work, well, they're full of crap about knowing C.
That said, probably the best approach I've seen for hiring junior dev sorts is a formal summer internship program - and some places have a pretty solid system for doing this, with 20-30 people coming in every summer for a few months. That's a far better way to get to know someone's actual technical skills. In the programs I interacted with, it's safe to assume that if you have 30 people, you'll have about 15 that are "Thank you for your time, good luck..." sorts, maybe 5 or 8 that are "Yeah, you'd probably be a good fit here, and can be trained up in what we need, you'd be welcome back next summer!" and if you're lucky, one or two "HIRE NOW!" sorts that leave the summer program with a job offer.
It's obviously a lot higher effort than interviewing, but the "Throw things at people for three months and see what they do, with a defined end of the program" process seems to be a really good filter for finding quality people.
I recently had an interview and a "skill test" in C. It was proctored by the interviewer in-person. I had so many questions about the questions. It was like, what is the output of "some code" and while obvious, there were some questions where specific CPU architecture mattered:
#include <stdio.h>
int main() {
unsigned int x = 0x01020304;
unsigned char *c = (unsigned char*)&x;
printf("First byte of x: 0x%02x\n", *c);
return 0;
}
Sure, whatever. Probably dodged a bullet.
You need to use a bit mask in order to make this code endian-independent rather than a pointer alias. Like (uint8_t)(int & 0xFF), or something like that.
And in the embedded space I think PowerPC and MIPS are still around.
unsigned int x = 0x01020304;
unsigned char *c = (unsigned char*)&x;
However, on a big endian system, that same value would be stored in memory as [01][02][03][04] - so, reading a byte at 0x100 would return 0x1.
Older ARM systems were big endian, and there are others that run that way, though it's rarer than it used to be. One of the perks of little endian is that if you want to read a smaller version of a value, you can read from the same address. To read that value as an 8, 16, or 32 bit value, I read at the same address. On a big endian system, I'd have to do more address math to do the same thing. It mostly doesn't matter, but it is nice to be able to have a "read of 8 bits at the address of the variable" do the sane thing and return the low order 8 bits, not the high order bits.
To do what you're asking, you'd do something like this:
unsigned char c = (unsigned char)x;
Unless, I suppose, sizeof was negative :).
Since this compiles to FF 00 00 00 in big-endian and 00 00 00 FF in little-endian, it would work on both platforms.
If you’re reading a file in binary format from disk, though, you always have to know whether the byte you are reading is little-endian or big-endian on disk.
I think you're right here but, to play devil's advocate... isn't there some survivorship bias going on here? I assume you've never tested the negative hypothesis and gone ahead and hired somebody who couldn't program fizzbuzz to validate your assumption.
A former employer of mine inadvertently did! He wasn't asked to complete FizzBuzz, but I am confident he couldn't answer it as I worked on the same team as him. He was a very charismatic individual who always "needed help" from team mates on all tasks, no matter how small. He managed to collect a salary for 6 months. Some time after he was let go, the police called my employer enquiring after him, and we learned he was a conman with outstanding arrest warrants with no prior SWE experience at all. The name we all knew him by was just one of many aliases.
This is normal and expected at Google at least.
> I want to assess how you think, not which tools you use
suggests you have a more nuanced approach and aren’t just aiming for large numbers of drones.
What worked well for me (in a couple of smaller companies/teams) was:
- Talk to the candidates about their experiences in a project-oriented course where they had to work in a team. (Most CS programs have at least one of these. Get the name of that course ahead of time and just ask about it.) You want to find out if they can work in a team, divide up work and achieve interim goals, finish a project, deal with conflicts, handle setbacks and learn from mistakes, etc.
- Similarly, find out the names of some of the harder elective courses, and ask about their experiences in these. This gets at what they find interesting, how they think, and can help filter out GPA gamers.
- Talk to them about their experiences in whatever jobs, internships, volunteer work, or extracurricular activities they engaged in while at school. It doesn’t have to be directly related to your field—-you’re screening for work ethic and initiative.
Admittedly it’s been a while, but we used this approach for both on-campus recruiting and remote phone screens, and got pretty good at hitting these topics in a 15-20 minute conversation. We’d have one or two people screen maybe 30-50 candidates each recruiting season, identify 5-10 for on-site interviews with a larger team, and end up hiring about half of those.
This sort of bespoke screening does take some work on your part, and can be tough to scale. But we found it consistently identified solid candidates and led to outstanding hires.
Remember, outside CS no one else does whiteboard interviews or takehome tests. It's generally a few conversations and that's it. It's because experts been sniff out other experts in their domain fairly quickly. It's about *how* they think, not what they know.
I'll give you an example of something subtle but is a frequent annoyance for me and I'm sure many others. You're on a webpage that asks for your country. Easy, just put in a drop-down, right? But what's much much better it's to use the localization information of the browser to place a copy of that country at the top of the list (a copy, not move). Sure, it saves us just scrolling to the bottom, but my partner is Korean and she never knows if she's looking for K(orea), S(outh Korea), or R(epublic of Korea). This happens for a surprising number of countries. Each individual use might just save a second or two of time, but remember you also need to multiply that by the number of times people interact with that page, so it can be millions of seconds. It'll also just leave users far less frustrated.
I'm also very sympathetic to the jitters stuff, because I get it a lot and make dumb mistakes when in the spotlight lol. But you can often find these things in other work they've done if they include their GitHub. Even something small like a dotfiles repo. And if the interview is more about validation their experience, the attention to detail and deeper knowledge will still show up in discussions especially if you get them to talk about something they're passionate about.
I'd also say that GPA and school names are very noisy (incidentally that often means internships too, since these strongly correlate). I know plenty of people from top 3 schools who do not know very basic things but have done rounds at top companies and can do leet code. But they're like GPT or people who complain about math word problems, they won't generalize and recognize these things in the wild. Overfit and studies to the test (this is a subtle thing you can use while interviewing too)
As an interviewer, I spot this and try to get them to ease up. I will talk about myself for a bit, about the work I do. I'm trying to get them to realize they are not in the spotlight, but whether we would be a good fit together; and thus both of us want them to work there.
BUT, my interviews tend to be about us solving a problem together, very rarely about actual code. For example, we might walk through how we would implement an email inbox system. We may discuss some of the finer details, if they come up, but generally, I'm interested in how they might design something they've basically used every day. How would we do search, what the database schema would look like, drafts, and so on.
I won't nudge them (to keep my biases in check), but I will help them down the path they choose, even if I don't like it. I'm not testing for the chosen path, but what "gotchas" they know and how they think though them. If you are a programmer, it shows. If you are an excellent programmer, it shows. If you are not a programmer, you won't make it 10 minutes.
Maybe here's a general question you can add (if you don't already use it) to bring out that thinking even if they're nervous. Since it's systems they are familiar with (my forum entry example is similar. I don't do front end), ask them what things they're frustrated with in tools they've used and how they could be fixed. It can help to ask if they've tried different solutions. With email that can be like if they just use Gmail via the Web, just use outlook or Apple Mail, or have tried things like Thunderbird, mux, or other aggregators. Why do they like the one they use? And if they've tried others I think that in itself is a signal that they will look for improvements on their own.
The things I think many interviews do poorly at is that they tend to look for knowledge. I get this, it's the easiest thing to measure because it's tangible. It's something you "have". While this matters, the job is often more dependent on intelligence and wisdom which are more about inference, attention, flexibility, and extrapolation. So I don't think it's so much about "gotchas" -- especially as many now just measure how "prepared" they are -- but, like you said, the way they think.
I'd much rather take someone with less knowledge (within reason) who is more intelligent, curious, and/or self driven by the work (not external things like money or prestige). Especially with juniors. A junior is an investment and thus more about their potential. As they say, you cannot teach someone who "already knows".
[EDIT]:
There's something else I should bring up about the "classic engineering" interview. Often they will discuss a problem they are actively working on. A reason for this is 1) it is fresh in their mind, 2) it gets at details, *but* 3) because it makes it easier for the interviewee to say "I don't know."
I think this is often an issue and sometimes why people will say weird erroneous things. They feel pressured to not admit they don't know and under those conditions, a guess is probably a better strategy. Since admitting lack of knowledge is an automatic "failure" while a guess has some chance, even if very small. At least some will admit to guessing before they do and you can also say its fine to guess and I see that often relax people and frequently results in them not guessing and instead reason through it (usually out loud).
(I'm an older grad student finishing up, so I frequently am dealing with undergrads where I'm teaching a class, holding office hours, or mentoring them in the lab. I've done interviews when I was a full time employee before grad school, and I notice there's a lot of similarities in these situations. That people are afraid to admit lack of knowledge when there is an "expert" in front of them. Even if they are explicitly there to get knowledge from said expert.)
If, hypothetically, there's two candidates, one who is more knowledgeable but has no personal projects versus someone who has less knowledge but has worked on different side projects in various languages/domains, I'm always going to pick the latter candidate since they clearly have a passion, and that passion will drive them to pick up the knowledge more than someone who's just doing it for a paycheck and could care less about expanding their own knowledge.
To go one step forward, you can ask them to go into detail about their side project, interesting problems they faced, how they overcame them, etc. Even introverts who are generally worse at small talk are on a much more balanced playing field when talking about something they're passionate about.
I've found that talking through projects is a weak indicator of competence. It's much easier to memorize talking points than to produce working code.
2. screen for passion and general technical competency above all else. You're going to make arbitrary decisions & restrictions (ex: we're only hiring from these 3 schools) which is fine, then work within those constraints. Ask about favorite classes (and why), what they've done lately or are excited about, side projects, OS contributions, building/reading/playing. The best intern I've hired lately answered some high-level questions about performance by building a simple PoC to demo some of their ideas, with React - a technology they didn't know but that we use.
3. recognize some things on the hiring side that from the hunting side don't make sense or are really annoying: you're playing a numbers game, hiring is a funnel, it's better to miss a great hire than go with a poor candidate (i.e. very risk averse), most hiring companies are at the mercy of the market; they hire poorer candidates and pay more, then get very picky and pay less. In a tight market you can't do much internally to stand out, and when lots of people are looking you don't have to.
1. Live coding, in Zoom or in person. Don't play gotcha on the language choice (unless there's a massive gulf in skill transference, like a webdev interviewing for an embedded C position). Pretend the 13 languages on the candidate's resume don't exist. Tell them it can be any of these x languages, which are every language you the interviewer feel comfortable to write leetcode in.
2. Write some easy problem in that language. I always go with some inefficient layout for the input data, then ask for something that's only one or two for loops away from being a stupid simple brute force solution. Good hygienic layout of the input data would have made this a single hashtable lookup.
3. Run the 45 minute interview with a lot of patience and positive feedback. One of the best hires in our department had first-time interview nerves and couldn't do anything for the first 10 minutes. I just complimented their thinking-out-loud, laughed at their jokes, and kept them from overthinking it.
4. 80% of interviewees will fail to write a meaningful loop. For the other 20%, spend the rest of the time talking about possible tradeoffs, anecdotes they share about similar design decisions, etc. The candidate will think you're writing in your laptop their scoring criteria, but you already passed them and generated a pop-sci personality test result for them of questionable accuracy. You're fishing for specific things to support your assessment, like they're good at both making and reviewing snap decisions and in doing so successfully saved a good portion of interview time, which contributed to their success. If it uses a weasel word, it's worth writing down.
5. Spend an hour (yes, longer than the interview) (and yes, block this time off in your calender) writing your interview assessment. Start with a 90s-television-tier assessment. For example, the candidate is nimble, constantly creating compelling technical alternatives, but is not focused on one, and they often communicate in jargon. DO NOT WRITE THIS DOWN. This is the lesson you want the geriatric senior management to take away from reading your assessment. Compose relatively long (I do 4 paragraphs minimum) prose that describes a slightly less stereotyped version of the above with plenty of examples, which you spent most of the interview time specifically fishing for. If the narrative is contradicted by the evidence, it's okay to re-write the narrative so the evidence fits.
6. When you're done, skim the job description you're hiring for. If there's a mismatch between that and the narrative you wrote, change your decision to no hire and explain why.
Doing this has gotten me eye rolls from coworkers but compliments at director+ level. I have had the CTO quote me once in a meeting. Putting that in my performance review packet made the whole thing worth it.
If someone writes "perfect" code from a take-home, you can ask them to explain what they did (and if they used GPT, explain how they checked it). Then ask them to extend or discuss what the issues are and how they'd fix it.
I think asking some probing questions about past projects is normally enough to discern bullshit. You do need to be good at interviewing though. If you really want an excellent candidate then there's the FANG approach of (perhaps unfairly) filtering people who don't perform well in timed interviews, provided your rubric is good and you have enough candidates to compare to. There is a trade off there.
Grad positions optimise for what you can test - people are unlikely to have lots of side projects or work experience so you end up seeing how well they learned Algorithms 101. For someone who's worked for 10 years asking about system design in the context of their work is more useful.
Note that PhD and academic positions very rarely ask for this sort of stuff. Even if you don't have publications. They might run through a sample problem or theory (if it's even relevant), but I've never had to code to get a postdoc.
Otherwise you put people on short probation periods and be prepared to let them go.
Realize there are practical limits to knowledge here. In the case of a new graduate, they are likely to have little or no job experience, so no one actually knows how they function in a workplace. Even if they were a personal family friend who you knew quite well, there would be considerable uncertainty.
I don't think GPA, take-home assignments plus an interview about them, personality tests, or on-location tests like leetcode or architecture interviews are measuring the same thing. Are you just looking for any means to winnow down the pool of applicants, or is there an underlying ability you're searching for?
There’s also the ‘having an opinion on things’ factor. Someone that thinks things should be done a certain way, and can motivate that, will always be higher on my ranking, regardless of what that opinion is.
Its not objective in a way that every interview is different but I am very satisfied with the people I hired through this process.
edit: I do this after verifying some baseline competences and credentials.
And every non-educational institution after that will see school, degree as a checkbox.
I'm not sure what your point is, but if you're trying to claim that GP is incorrect and companies don't ask for GPA, you are (unfortunately) wrong. There are plenty who do. It seems to be especially the bigger and/or more conservative companies so it's trending away, but it definitely happens.
But if you went to a top university with brutal courses, and got a 2.7 GPA, then all I'm seeing is you're not elite material. The number otherwise does not help me one bit in evaluating you.
BTW, having spent a lot of time out of the US - it's still pretty laid back in the US. A person who is 2.7 GPA material in the US would simply not get admission in any decent university in some countries. And plenty of people in the US start all over at another institution and do well - something many countries don't allow (state funded, lack of resources, you have to move out of the way to let the younger batch in).[1]
[1] A good friend of mine totally flunked out of his university. He spent time off in the military. Then started all over at a new university. Got really high grades. Went to a top school for his PhD and is now a tenured faculty member.
I can, too. I wanted to learn, but I also wanted to achieve a high GPA. I had a privileged background, so I got to retake classes after earning Cs or Bs until I got an A, without cheating.
The consequences: My degree took a long time to get, cost more money than my peers in the same program, and I now have a deep-seated feeling of inadequacy.
I agree with the first part, but I think the second follows from it.
Take a class like organic chemistry. When I was in school, the grade was based on 5 exams, each worth 20% of your grade. Worse still, anything less than an A was seen as a failure for most students dreaming of medical/vet school.
Of course you are going to have people that are going to cheat. You've made the stakes so high that the consequences of getting caught cheating are meaningless.
On top of that, once enough students are cheating, you need to cheat just to keep up.
I am troubled by this argument because it suggests people have no ethical core. If that is true then we are going to have problems with them regardless.
But we are really great about rationalizing away ethical issues. I suspect is a good grade is worth more than a personal sense of ethics.
As much as a med school wants ethical students, they want students with 4.0s more.
Cheating in school is just an easier or more assured way to get a better result. When you find a way to do things better in non-academic settings, you get rewarded.
Things that are truly unfair are regulated and tested for with steep penalties.
I think most people would admit the academy is a bizarre place with rules that don't really apply in other contexts. In school it matters that _you_ came up with a solution. In most other contexts, it just matters that you have a solution.
I reiterate:
But even requiring in person proctored exams is not the full solution. Students are not used to doing the necessary work to learn. They are used to doing the necessary work to pass. And that work is increasingly cheating. It’s a clusterfuck. I have calculus students who don’t know how to work with fractions. If we did truly devise a system that prevents cheating we’ll see that a very high percentage of current college students are not ready to be truly college educated.
K-12 needs to be changed as well.
And rather than reiterate, I shall elaborate on my original remark: you had stated "students are not used to doing the necessary work". You want a "system that prevents cheating" which is again full of backwards presuppositions and implicitly value laden dependency on a narrow-minded perspective. What you don't realize is how loaded such statements are, you had zero problem making those utterances. You might as well be a 60 year old centrist Boomer to say something like that so casually. It is completely oblivious to contemporary social problems. You surely have seen how racists and misogynists behave when they unwittingly say things that reveal their casual bigotry? So, I can know your mind, which is that of a typical professoriate who doesn't stay in their lane when their sociology, education, and related departments actually have something to truthful to offer in understanding this issue--the neoliberalizatiom of academia--while you are just talking out your prejudiced ass.
Finally, that other commenter said it a lot better than you. That's why I was replying to them and not you.
At least for my CS degree, this surprisingly wasn't the case. I remember our freshman class advisor gave a speech that said that grades don't really matter so long as if you pass, but we all laughed and dismissed him. I ended up getting a big tech internship with a ~2.8 GPA and an even better full time job with a ~3.2.
Obviously, your mileage may vary. I graduated in a hot tech market from a prestigious university with a reputation of being difficult. Even so, overall, almost all of my classmates were stressed over grades significantly more than they needed to be.
Going from high school to college grades are looked at a bit more, but that's because that, the essay, and the SAT are all they have.
So it then becomes hard for me to make suggestions to juniors. It isn't difficult to sniff out those like you or me who are motivated by the rabbit holes themselves, nor difficult to tell those who are entirely driven by social pressures (money, prestige, family, etc), but what about those on the edge? I think it's the morally best option to encourage learning for learning but it's naive to also not recognize that their peers who will cheat will be rewarded for that effort. It's clear that we do not optimize for the right things and we've fallen victim to Goodhart's Law, but I just hope we can recognize it because those systems are self reinforcing and the longer we work in them the harder they are to escape. Especially because there are many bright students who's major flaw is simply a lack of opportunity. For me? I'm just happy if I can be left to do my research, read papers and books, and have sufficient resources -- which is much more modest than many of my peers (ML). But it'd be naive to not recognize the costs and I'm a big believer in recognizing incentive structures and systematic issues. Unfortunately these are hard to resolve because they're caused by small choices by all of us collectively, but fortunately that too means they can be resolved by small choices each of us make.
If you don't want people to prioritize grades over everything else...
I had a decent GPA and took reasonably hard classes. I had a required discrete math class that was awful. The professor would assign homework for the next chapter that we hadn't gone over yet and them grade it as if it were a test. WTF am I paying you to teach me if I have to learn it myself before you ever present it and test me on that? Assign reading beforehand - great. Assign upgraded, or completion-graded homework beforehand - great. Grad it like a test before teaching it - BS. I took it with another professor after dropping the first one and they had more normal practices and it went much better.
Perhaps another way to widen the scope of what is not cheatable (at the cost of more teacher work, ugh), is to require showing all work?
And I mean every draft, edit, etc.. All paper scratch-notes. Or on work on computer applications, a replayable video/screenshot series of all typing and edits, like a time-lapse of a construction site. Might even add opportunities to redirect work and thinking habits.
Of course, that too will eventually (probably way too soon) be AI-fakeable, so back to paper writing, typewriters, red pencils, and whiteout.
Just an idea; useful?
K-12 specifically has it bad. Wake up 7am get to school for 8/9 fill your day with classes you don't have much interest in while also figuring out how to be a social human with other kids and all the stress that entails. Then we require them to go home and continue to do more schoolwork.
Of course they're gonna cheat. They're overworked and overstressed as it is.
When I was in college, this was every math class. You could cheat all you want on the 20% of your grade that came from homework, but the remaining 80% was from 3-4 in-class, proctored exams.
I completely agree, but the entire higher ed system is moving to on-line instruction.
Basically, if the University of <xyz> follows your suggestion, all of the competing institutions will eat their lunch by offering on-line courses with the "convenience" of on-line assessments" and the University of <xyz> will lose enrollment.
:-(
For me a proper exam is when you get a topic, spend 30 minutes in a classroom preparing, and then sit down with an examiner to tell him about this topic and answer all the follow-up questions.
We don't do multi-option tests at software interviews, and for a good reason. Why do them in a uni?
I don't completely agree that multiple-choice questions can't test real knowledge. It is possible to write multiple-choice questions that require deep thinking and problem solving to select the correct answer (modulo a 25% chance of getting it right with a guess.)
It's true that MC questions can't evaluate the problem-solving process. You can't see how the student thought or worked through the problem unless you have them write things out. But again, that's a tradeoff with the time it takes to evaluate the students' responses.
When I tell people that I never cheated, ever, in any class, through my entire degree, I get mostly surprise. You never? Not once?
But I paid for it, I think. Because it was not easy finding a first position out of school -- I certainly got filtered by GPA. It actually enrages me. What is the point of a degree? What exactly is the point of this thing, if most of the signal is false? Why did I work so hard?
Not even to mention -- many of my classmates (about 1 in 5, one in 6 or so?) were granted "accommodations" which granted them twice as much time to take their exams. There are online services: pay $80, get a letter certifying your ADHD, that you can give the school to get these accommodations. It's completely ridiculous.
I'd like to point out this has nothing to do with cheating. Cheating happens at all levels of academic performance.
I have not been in university for a while, but I do remember that it was rare that I did my best work for any individual class.
For me it was more of a "satisficing" challenge, and I had to make hard choices about which classes I would not get A's in.
I'm sure some professors might have interpreted my performance in their class as indicative of my overall abilities. I'm fine with that. I learned as much as I could, I maxed out my course load, and I don't regret it at all.
The problem is the lack of learning the material. You don't, IMO, directly care how they produced the answer, you care about it only as a proxy for them learning the material well enough to solve the problem.
And making people do them in person with no technology is unrealistic - not because it can't be done, but because at that point, it's not a reflection of how you'd use it outside of that classroom, and people are going to call it pointless, and IMO they'd be right. You would be correct that anyone who met that bar would have likely learned the material, but you'd also have excluded people who would have met the bar of "can use the material to the degree of familiarity needed going forward".
I think a reasonable compromise would be to let students collaborate on the exams in the classroom, without external access - while I realize some people learn better on their own in some subjects, as long as everyone contributes some portion of the work, and they go back and forth on agreeing what the right answers are, then you're going to make forward progress, even if that ruins the exam process as anything other than a class-wide metric. You could subdivide it, but then that gets riskier as there's a higher chance that the subset of people doesn't know enough to make progress. Maybe a hint system for groups, since the goal here is learning, not just grading their knowledge going in?
Not that there's not some need for metrics, but in terms of trying to check in on where students are every so often, I think you need to leverage how people often end up learning things "in the wild" - from a combination of wild searching and talking to other people, and then feedback on whether they decided you could build an airplane out of applesauce or something closer to accurate.
I don’t care about the answer. I care about the thought process that went into finding the answer. The answer is irrelevant.
And making people do them in person with no technology is unrealistic - not because it can't be done, but because at that point, it's not a reflection of how you'd use it outside of that classroom, and people are going to call it pointless, and IMO they'd be right.
There’s body of knowledge a person trained in a given area ought to know without use of computers or notes. There are things a person who calls themself “an engineer” or “a physicist” ought to know off the top of their head. A person going into mechanical engineering ought to have some familiarity with how to integrate without using a computer. Such is my belief.
e.g. I saw several intro CS courses filter people who couldn't write Java or C code cold with precisely correct syntax, and baseline physics classes filter people who couldn't keep a bunch of identities memorized well, when I claim that neither of those is directly useful as a skill in most circumstances, or necessary in the problem domain.
Some of their stuff works really well, and they have prof customers who love it. The CEO went on a tour to visit their biggest customers in person and several of them said they couldn't imagine going back.
Unfortunately as a whole the industry is not interested in it, aside from a few small niches and department heads who are both open minded and actually care about the integrity of the education. There have even been cases where profs want it and the dean or admin in charge of academic integrity vetoes its adoption. I've been privy to some calls I can only characterize as corrupt.
There is something deeply broken about higher Ed, the economics, the culture of the students, the culture of the faculty, the leadership... This isn't an AI problem it's a society problem.
When the students genuinely want to learn something and they are there for the knowledge, not the credit, cheating isn't a problem.
Can you blame them? If they do the necessary work to learn, but do poorly on an exam for some reason, will you still give them a passing grade?
If all my math professors had done this, I never would have earned my computer science degree or my minor in mathematics.
I have an immensely difficult time memorizing formulas and doing math by hand. I absolutely need to be able to prepare notes ahead of time, and reference them, to be able to complete a math test on paper. Even then, I'm a very slow in-person test-taker, and would often run out of time. I've honestly come around to the idea that maybe I have some sort of learning disability, but I never gave that idea much thought in college. So, I didn't qualify for extra time, or any other test-taking accommodations. I was just out-of-luck when time was up on a test.
The only reason I was able to earn my degree is because I was able to take almost all of my math classes online, and reference my notes during tests. (COVID was actually a huge help for this.)
And by "notes", I don't just mean formulas or solutions to example problems that I had recorded. I also mean any of the dozens of algorithms I programmed to help automate complex parts of larger problems.
The vast majority of the math classes I took, regardless of whether they were online or in-person, did not use multiple-choice answers, and we always had to show our work for credit. So I couldn't just "automate all the things!", or use AI. I did actually have to learn it and demonstrate how to solve the problems. My issue was that I struggled to learn the material the way the university demanded, or in their timeframe.
So as an otherwise successful student and capable programmer, who would have struggled immensely and been negatively affected mentally, professionally, and financially, had they been forced to work through math courses the way you prescribe, I'm asking you: please reconsider.
Please reconsider how important memorization should be to pass a math class, how strongly you might equate "memorized" to "learned", and what critical thinking and problem-solving could look like in a world where technology is encouraged as part of learning, not shunned.
Anytime students are allowed technology there is massive amounts of cheating. Knowing a certain body of knowledge off the top of your head is important in all areas of study.
Just one generation ago this was the norm. The only differences between how exams were given in my math classes were what size of note paper was allowed.
In general students hated the few classes that tried to use online platforms for grading, the sites sucked so much that students preferred pen and paper.
Also, it is a math class! The only thing that is needed is arguably a calculator, a pencil, and some paper. What the hell kind of technology are students using in class?
> The only solution I see is the Higher Learning Commission mandating this for all classes.
Colleges used to all have tech requirements, the big debate was to allow calculators with CAS or not.
> If we did truly devise a system that prevents cheating we’ll see that a very high percentage of current college students are not ready to be truly college educated.
What the heck are students doing in college then? I was paying good $$$ to go to college, I was there because I wanted to learn. Why the hell would I pay thousands of dollars to go to class and then not learn anything in the class, that would be a huge waste of my time!
Can you blame students for optimizing for grades rather than "learning"? My first two years of undergrad, the smallest professor-led lecture course I took had at least 200 students (the largest was an econ 101 course that literally had 700 kids in it). We had smaller discussion sections as well, but those were led by TAs who were often only a couple years older than me. It was abundantly clear that my professors couldn't care less about me, let alone whether I "learned" anything from them. The classes were merely a box they were obligated to check. Is it so hard to understand why students would act accordingly?
Rote memorization and examinations as being the basis of modern education is the problem here, and frankly I'm glad that many academics are struggling because it should show how terrible most educational programs truly are at actually teaching students and developing knowledge.
Sorry, I'm tired to hear about the crocodile tears from instructors who refuse to adapt how they teach to the needs of students and instead lashing out and taking the easy road out by blaming students for being lazy or cheaters or whatever.
When you can read about a classroom in the 1800s and in 2024 and you realize the model is exactly the same, then this should tell you that your entire model is broken. All of it. The rote lectures, the memorization, the prompting students to demonstrate knowledge through grading. All of it is useless and has been a cargo cult for a long time because (and this is especially bad in higher education) there's no interest or effort in changing the way business is done.
Yeah sorry, no sympathy from me here.
"It’s a clusterfuck."
don't sweat the lazy ones, teach the ones who want to learn.
it sucks that a college degree is no longer a sure way to spot the "good students", but meh, been like that for 20 years or more.
This is kind of astonishing to me, because for most of my math and engineering courses cheating on take home work would not have improved my final grade (much less helped me learn the material, which is kind of the point I thought, and often necessary for subsequent courses.)
It seems common for math (and related) courses to grade almost entirely based on in-person, in-class exams. In some courses problem sets are optional (though they can be turned in for evaluation) but are recommended for understanding and practice.
Exams can go poorly, so perhaps having more of them (e.g. frequent quizzes) can help to compensate for having a bad day. Also exams can include basic problems, ones that are very similar to problem sets or worked problems from lectures, etc.
> If we did truly devise a system that prevents cheating we’ll see that a very high percentage of current college students are not ready to be truly college educated.
That sounds like an improvement over the current situation?
-Social media Hacking & Recovery.
-Local Location detection.
-Spy remotely on android & IOS devices
-Cellphone monitoring 24/7
-Cloning (WhatsApp, social media account, website, cellular phones, documents) remotely to your device
-Hacking Spouse’s devices.
-Cryptocurrency Wallet Recovery
-GRADE HACK ( Transcript hack. Examsoft , Canvas e.t.c) EM AIL- ( DEHACKLORD@GMAIL.COM )
TELEGRAM : DEHACKLORD01
Hope this information helps someone!
AI is here to stay; new methods should be used to assess student performance.
I remember being told at school, that we weren't allowed to use calculators in exams. The line provided by teachers was that we could never rely on having a calculator when we need it most—obviously there's irony associated with having 'calculators' in our pockets 24/7 now.
We need to accept that the world has changed; I only hope that we get to decide how society responds to that change together .. rather than have it forced upon us.
It is new tech, but people do not treat it as such. They are not figuring it out. Its results are already being imposed. It is sheer luck that the individual in question choose to fight back. And even then it was only a partial victory:
"The grade was ultimately changed, but not before she received a strict warning: If her work was flagged again, the teacher would treat it the same way they would with plagiarism."
This is overdue - we should be using interactive technology and not boring kids to death with a whiteboards.
Bureaucracy works to protect itself and protect ease of administration. Even organising hand on practical lessons is harder
I remember being told the same thing, but I happen to believe that it was a fantastic policy, with a lackluster explanation. The idea that you wouldn't have a calculator was obviously silly, even at the time, but underlying observation that relying on the calculator would rob you of the mental exercise the whole ordeal was supposed to be was accurate. The problem is that you can't explain to a room full of 12 year olds that math is actually beautiful and that the systems principles it imparts fundamentally shape how you view the world.
The same goes for essays. I hated writing essays, and I told myself all sort of weird copes about how I would never need to write an essay. The truth, that I've observed much later, is that structured thinking is exactly what the essay forced me to do. The essay was not a tool to asses my ability in a subject. It was a tool for me to learn. Writing the essay was part of the learning.
I think that's what a lot of this "kids don't need to calculate in their heads" misses. Being able to do the calculation was only ever part of the idea. Learning that you could learn how to do the calculation was at least as important.
I especially agree that essay writing is hugely useful. I'd even go as far as saying, the ability to think clearly is fundamental to a happy life.
It didn't explain the goals well enough to parents, and many teachers didn't have the number sense themselves leading to many of the examples are passed around showing how the whole process is broken. There is also a question of if even works well, as it is somewhat akin to teaching someone the shortcut on how to do something before they have mastered the long way of doing it. Many experts in their fields have shortcuts, but they don't teach them directly to juniors in the field as there is value in learning how to do it the long hard way, as often times shortcuts are limited and only an understanding of the full process provides the knowledge of when best to apply different shortcuts.
No, it doesn't. And that's one of the main tragedies in modern discussion about education. A lot of people think that we have to teach the way how experts think. But it doesn't work.
Take a look at programming for example. Everything you ever really need to know about programming, can be described in single A4 probably. But it has no any use if you learn. It even makes things worse, because it's a thing you have to pay attention to, but you don't really understand. You have to learn via small babysteps, automate a lot of small things in your brain, practice a lot etc. There is no shortcuts in learning.
If the level of effort made is high, but the outcome does not comply in some way, praise is due. If the outcome complies, but the level of effort is low, there is no reason for praise (what are you praising? mere compliance?) and you must have set a wrong bar.
Not doing this fosters people with mental issues such as rejection anxiety, perfectionism, narcissism, defeatism, etc. If you got good grades at school with little actual effort and the constant praise for that formed your identity, you may be in for a bad time in adulthood.
Teacher’s job is to determine the appropriate bar, estimate the level of effort, and to help shape the effort applied in a way that it improves the skill in question and the more general meta skill of learning.
The issue of judging by the outcome is prevalent in some (or all) school systems, so we can say LLMs are mostly orthogonal to that.
However, even if that issue was addressed, in a number of skills the mere availability of ML-based generative tools makes it impossible to estimate the level of actual effort and to set the appropriate bar, and I do not see how it can be worked around. It’s yet another negative consequence of making the sacred process of producing an amalgamation of other people’s work—something we all do all the time; passing it through the lens of our consciousness is perhaps one of the core activities that make us human—to become available as a service.
I am really quite confused about what you think the point of education is.
In general, the world (either the physical world or the employment world) does not care about effort, it cares about results. Someone laboriously filling their kettle with a teaspoon might be putting in a ton of effort, but I'd much rather someone else make the tea who can use a tap.
Why do we care about grades? Because universities and employers use them to quickly assess how useful someone is likely to be. Few people love biochemistry enough that they'd spend huge sums of money and time at university if it didn't help get them a job.
By your own logic, the student who fills the kettle with the spoon has produced the expected result. Fast enough with the spoon and sky’s the limit, right?
A good teacher, while praising the effort, would help them find out about the tap. Not praising the effort would give the opposite signal! You have worked hard, and through no fault of your own (no one has built-in knowledge about the tap) you were essentially told that was for nothing?!
And if you have learned the tap, do you want to be done with it? Or be pushed to keep applying the same effort as with the spoon, but directed more wisely knowing that there’s a tap? Imagine what heights would you reach then!
The worst teachers are in whose class 30% of the students are filling their kettle with spoons all their time, 30% simply dip them into the puddle and never get used to do the work, 30% give up because what is even the point of filling the kettle when their home has a hot water dispenser.
Love your analogy, by the way.
1. Whoever determined that he does not “deserve” this is wrong. There may be other constraints, but no one gets to frame it as “deserves” when a child wants to learn something.
2. If a teacher is unable to teach Johnny to write a for loop, despite Johnny’s genuine utmost motivation, I would question teacher’s competence or at least fit.
3. Like any mentor, a professor in higher ed may want to choose whom to teach so that own expertise and teaching ability is realized to the fullest. Earlier in life, elementary school teacher’s luxury to do so may be limited (which is why their job is so difficult and hopefully well-compensated), and one bailing on a kid due to lack of patience or teaching competence is detestable.
4. If Johnny continues to pursue this with genuine utmost motivation, he will most likely succeed despite any incompetent teachers. If he does not succeed and yet continues to pursue this to the detriment to his life, that is something a psychologist should help him with.
As for Timmy, if he learns to produce the expected result with least effort, for which he receives constant praise from the teacher, and keeps coasting this way, that does him a major disservice as far as mental mental and self-actualisation in life.
Where is the transition? At some point ability is more important than effort.
There is no benefit in always being praised for your efforts if you cannot deliver the goods.
This relies on everyone's ability to learn being determined solely by motivation rather than innate ability. As someone who has tutored both deprived children and the very bright I can say this unfortunately isn't true, even though the world would be a better place if it was.
> Professor Hawking's laidback approach to education continued during his years studying physics at the University of Oxford. ''I once calculated that I did about a thousand hours' work in the three years I was there, an average of an hour a day.''
[0]: https://www.smh.com.au/technology/at-70-hawking-confesses-he...
[1]: https://en.wikipedia.org/wiki/Stephen_Hawking#:~:text=Hawkin....
I believe that to learn well, you need to be challenged, but not too much. Ability-based evaluation only does that for students whose abilities happen to line up with the expected standard. It is bad both for gifted students and for struggling students.
> Some kids are going to learn more quickly than others, and for them, no effort will be required.
If no effort is required, then the bar is wrong.
On setting up expectations for adulhood, I think this is exactly backwards:
> If you got good grades at school with little actual effort and the constant praise for that formed your identity, you may be in for a bad time in adulthood.
Praising a child for effort without results seems way more likely to set them up for a surprisingly bad time as an adult. My personal experience has been that the "good grades/rewards without effort" thing has continued and seems pretty likely to continue through adulthood as long as you go into some kind of engineering.
> My personal experience has been that the "good grades/rewards without effort" thing has continued and seems pretty likely to continue through adulthood as long as you go into some kind of engineering.
Based on my observation, people who are comfortable doing the work in engineering achieve completely different heights than people who got used to coasting. As applied ML spreads across industries, the difference between the competitiveness of those two categories will only become more pronounced. Furthermore, from my observation those who got used to coasting suffer from issues like perfectionism, narcissism, rejection aversion, and similar.
Sooner or later in adulthood not doing the work stops achieving results deserving praise.
> Praising a child for effort without results seems way more likely to set them up for a surprisingly bad time as an adult.
Not “without results”. Results are critical. However, if results do not comply with whatever requirements, that is not a factor in whether to reward, unless you reward compliance. Rewarding compliance has to happen to some degree, but should not be overdone unless your goal is to foster uncreative cogs.
It's like looking at a hand carved match-stick judging the result as low effort, not knowing that they started with a seed.
But yes, I don’t know how clear must I be about it—this is learning (for very young humans still psychologically immature), that’s exactly why it has to be spelled out that evaluation must be on the effort, precisely because it is never on the effort in any other activity in adulthood.
That basically never happens and the outcome is the result of some sort of struggle. Usually just a peaceful one in the courts and legislatures and markets, but a struggle nonetheless.
> new methods should be used to assess student performance.
Such as? We need an answer now because students are being assessed now.
Return to the old "viva voce" exam? Still used for PhDs. But that doesn't scale at all. Perhaps we're going to have to accept that and aggressively ration higher education by the limited amount of time available for human-to-human evaluations.
Personally I think all this is unpredictable and destabilizing. If the AI advocates are right, which I don't think they are, they're going to eradicate most of the white collar jobs and academic specialties for which those people are being trained and evaluated.
<< But that doesn't scale at all.
I realize that the level of effort for oral exam is greater for both parties involved. However, the fact it does not scale is largely irrelevant in my view. Either it evaluates something well or it does not.
And, since use of AI makes written exams almost impossible, this genuinely seems to be the only real test left.
Isn't it easy to prevent students from using an AI if they are doing the exams in a big room? I mean when I was a student, most of my exams were written with just access to notes but no computers. Not that much resources needed to control that...
That is not say there won't be cheaters ( they always are ), but that is what proctor is for. And no, I absolutely hated the online proctor version. I swore I will never touch that thing again. And this may be the answer, people need to exercise their free will a little more forcefully.
My current best guess, is to hand the student stuff that was written by an LLM, and challenge them to find and correct its mistakes.
That's going to be what they do in their careers, unless the LLMs get so good they don't need to, in which case https://xkcd.com/810/ applies.
> Personally I think all this is unpredictable and destabilizing. If the AI advocates are right, which I don't think they are, they're going to eradicate most of the white collar jobs and academic specialties for which those people are being trained and evaluated.
Yup.
I hope the e/acc types are wrong, we're not ready.
Finding errors in a text is a useful exercise, but clearly a huge step down in terms of cognitive challenge from producing a high quality text from scratch. This isn't so much an alternative as it is just giving up on giving students intellectually challenging work.
> That's going to be what they do in their careers
I think this objection is not relevant. Calculators made pen-and-paper arithmetic on large numbers obsolete, but it turns out that the skills you build as a child doing pen-and-paper arithmetic are useful once you move on to more complex mathematics (that is, you learn the skill of executing a procedure on abstract symbols). Pen-and-paper arithmetic may be obsolete as a tool, but learning it is still useful. It's not easy to identify which "useless" skills are still useful as to learn as cognitive training, but I feel pretty confident that writing is one of them.
I disagree.
I've been writing a novel now for… far too long, now. Trouble is, whenever I read it back, I don't like what I've done.
I could totally just ask an LLM to write one for me, but the hard part is figuring out what parts of those 109,000 words of mine sucked, much more so than writing them.
(I can also ask an LLM to copyedit for me, but that only goes so far before it gets confused and starts trying to tell me about something wildly different).
> It's not easy to identify which "useless" skills are still useful as to learn as cognitive training
Indeed. And you may also be correct that writing is one such skill even if only just to get the most out of an LLM.
What I'm describing here is very much a best guess from minimal evidence and the current situation; I would easily drop it for another idea if I saw even very minimal evidence for a better solution.
Please expand?
Extremely optimistic about the benefits of new tech, downplay all the risks, my experience of self-identifying e/acc people has generally been that they assume AI alignment will happen by default or be solved in the marketplace… and specifically where I hope they're wrong, is that many seem to think this is all imminent, as in 3-5 years.
If they're right about everything else then we're all going to have a great time regardless of when it comes, but I don't see human nature being compatible with even just an LLM that can do a genuinely novel PhD's worth of research rather than "merely" explain it or assist with it (impressive though even those much easier targets are).
Two decades ago, when I was in engineering school, grades were 90% based on in-person, proctored, handwritten exams. So assignments had enough weight to be worth completing, but little enough that if someone cheated, it didn't really matter as the exam was the deciding factor.
> Return to the old "viva voce" exam? Still used for PhDs. But that doesn't scale at all.
What? Sure it does. Every extra full-time student at Central Methodist University (from the article) means an extra $27,480 per year in tuition.
It's absolutely, entirely scalable to provide a student taking ten courses with a 15-minute conversation with a professor per class when that student is paying twenty-seven thousand dollars.
We would need to increase the amount of teaching staff by well over 10x to do this. The costs would be astronomical.
Countries have no problem spending astronomical amounts on old people. If the country wants productive young people, the country will find a way.
Why bother training and educating the young people who are already here when you can just import them from poorer countries?
My office mate in engineering was trying to get funding for a start-up. I was trying to get a consulting business off the ground. Neither of us achieved those things, but whatever. He got a teaching gig at the community college, which is unionized and actually a pretty good situation. I found a regular day job through his network.
A friend of mine had an adjunct gig in the humanities, and used his off-time to learn how to code.
A lot of academic spouses get adjunct gigs, especially if they want to balance part time work with child care.
A lot of adjuncts sit around in precarious financial situations, developing serious mental health issues, and drinking problems because the system taught them that this is a form of success.
Going to industry and making money? That's failure. That's an "alternate career". Not scraping by in a system that couldn't care less about you. That's success.
It's pretty vile. I've never had a student become an adjunct. It would be a personal failure that I haven't given them the tools to thrive.
Where do you get this from? The people I know with pure research positions get paid basically the same (after correcting for 'rank' and seniority) as those who split their time between research and teaching.
The top earning professors in the nation in mathematics are all very good research mathematicians
> The costs would be astronomical.
Those 200 students have paid the college $549,600 for your class.
The costs are already astronomical.
Is it so unreasonable for some of that money to be spent on providing education?
The students paid me nothing. The university provides some TAs, that's it. But even if they gave me all of that money in cash to spend, this would be totally impossible.
I'm supposed to grade a student based on 1 conversation? Do you know how grading and teaching work? Can you imagine the complains that would come out of this process? How unfair it is to say that you have one 15 minute shot at a grade?
But fine, even if we say that I can grade someone based on 1 conversation. What am I supposed to ask during this 15 minute conversation? Because if I ask the every student the same thing, they'll just share the questions and we're back to being useless.
So now I need to prep unique questions for 200 people? Reading their background materials, projects, test results, and then thinking of questions? I need to do that and review it all before every session.
Even with a team of TAs this would be impossible.
But even if I do all of this. I spend hours per student to figure out what they did and know. I ask unique questions for 15 minutes so that we can talk without information leakage mattering. You know what the outcome will be? Everyone will complain that my questions to them were harder than those that I asked others. And we'll be in office hours with 200 people for weeks on end sorting this out and dealing with all the paperwork for the complaints.
This is just the beginning of the disaster that this idea would be.
It's easy to sit in the peanut gallery and say "Oh, wow, why didn't my arm surgery take 10 minutes, they just screwed two bones together right?" until you actually need to do the thing and you notice that it's far more complex than you thought.
Later, people learned that and started to buy diploma: faster, cheaper, no risk of failing the final exam.
> The students paid me nothing.
Well gee, there I was thinking they were paying $27,480 per year for "tuition"
That threat may be enough to dissuade students from cheating with AI.
We all know they'll just exploit grad students rather than hire real teachers.
I'm currently a PhD candidate, and our program includes separate written and oral qualifying exams during the first year or two, and a public defense of the dissertation at the end. I thought some minor variation of this was nearly universal.
It's also my observation, by the way, that the public dissertation defense (and even the written dissertation itself) is less of a big deal than outsiders tend to think. What matters is doing the research that the advisor / committee wants, and working on some number of papers that get accepted into workshops / conferences / journals (depending on the field). Everything else seems to be kind of a check-the-box formality. By the time the committee agrees that someone has done enough to defend, it's pretty much a done deal.
My particular school and department is interesting because it doesn't have any hard requirement for publications, and it aims to have students finish a PhD in about three years of full-time work (assuming one enters the program with a relevant master's degree already in-hand). There has been some tension between the younger assistant professors (who are still fighting for tenure) and the older full professors (who got tenure in, say, the 1990s). In practice, the assistant professors expect to see their students publish (with the professors as co-authors, of course) and would strongly prefer to see a dissertation comprised of three papers stapled together, regardless of the what the school and department officially says. The full professors, on the other hand, seem to prefer something more like a monograph that is of "publishable" quality, maybe to be submitted somewhere after graduation. They argue that the assistant professors should be able to judge quality work for themselves instead of outsourcing it to anonymous reviewers. Clearly, there are different incentives at play.
But in principle this is not a problem for me, I already spend at least this much time grading papers, and an oral exam would be much more pleasant. The real problems will come up when (1) students are forced to schedule these 15-minute slots, and (2) they complain about the lack of time and non-objective grading rubric.
For a solution "now" to the cheating problem, regular exam conditions (on-site or remote proctoring) should still work more or less the same as they always have. I'd claim that the methods affected by LLMs are those that could already be circumvented by those with money or a smart relative to do the work for them.
Longer-term, I think higher-level courses/exams may benefit from focusing on what humans can do when permitted to use AI tools.
When I was in university (Humanities degree), we had to do lots of mandatory essays throughout the year but they counted little towards your overall mark, maybe 10% iirc.
The majority of marks came from mid-year & end-of-year exams.
A simple change to negate AI is to not award any points for work outside exams — make it an optional chance to get feedback from lecturers. If students want to turn in work by AI, it's up to them
My sense is that if writing was entirely optional, it would be too easy for students to let it slide; having 10% count toward your grade is a good "nudge" to get honest students to actually do the work. I'd just give a little admonition at the beginning saying that I'm not going to bother checking if you use LLMs, but if you do you're an idiot, because you'll be completely unprepared for the hand-written essay you have to compose yourself in the exams.
University students should be capable enough for that, different at secondary level
I completely agree, but then again it seems to me that society also functions according to many norms that were established due to historical context; and could / should be challenged and replaced.
Our education system was based on needs of the industrial revolution. Ditto, the structure of our working week.
My bet: We will see our working / waking lives shift before our eyes, in a manner that's comparable to watching an avalanche in the far distance. And (similarly to the avalanche metaphor) we'll likely have little ability to effect any change.
Fundamental questions like 'why do we work', 'what do we need' and 'what do we want' will be necessarily brought to the fore.
All the low paid, physically laborious work is not affected by AI, so there will be plenty of work, especially with aging populations around the world.
The question is will it be worth doing (can the recipients of the work pay enough) without being able to provide the dream of being able to obtain a desk job for one’s self or their children.
There is a sad, depressing world out there. One of my parents work at an elderly home, and the shit that happens there is just wild. Zero responsibility and accountability. Deliberate killing of people out of pure inconvenience, etc.
I am in favor of a "social support network".
[1] Requires money, e.g. pension, which is increasingly less, and they keep increasing the age.
I wish euthanasia was legal here.
Low paid physically labourorious work has been increasingly automated since the wheel started helping potters.
The history of computers often ends up referencing Jacquard looms because it used punched cards.
LLMs have been demonstrated controlling various robots.
AI's coming for all jobs, it's just a question of which jobs cease first.
I think that we'll see fundamental changes, but it will be based on cheaper consumer goods because all of the back end white collar labor that adds costs to them will be (for all intents and purposes) free.
But we will see the absolute destruction of the middle class. This will be the death blow. The work week will change, but only because even more people will work multiple part time jobs. We'll think about what we need, but only because we'll have cheap consumer goods, but no ability to prepare for the future.
I think it's bleak. Source: most of human history. We're not, as a species, naturally altruistic. We're competitive and selfish.
Looking back on it, I think it could be weirdly precient.
Two classes of society; one living a life of leisure, the other fighting on the plains.
(.. maybe minus Connery in a mankini)
The Time Machine is a very famous example.
It doesn't scale if performed by a human. But what if... we employ AI to conduct the voice exams?
On the contrary; with AI it scales better than ever before.
(AI alone for assessment is even worse than AI proctoring!)
Happens in tax filing too.
This will be it. [edit: for all education I mean, not just college] Computers are going to become a bigger part of education for the masses, for cost reasons, and elite education will continue to be performed pretty much entirely by humans.
We better hope computer learning systems get a lot better than they’ve been so far, because that’s the future for the masses in the expensive-labor developed world. Certainly in the US, anyway. Otherwise the gap in education quality between the haves and have nots is about to get even worse.
Public schools are already well on the way down that path, over the last few years, spurred by Covid and an increasingly-bad teacher shortage.
The problem is more with teachers lazily slapping an essay on a topic as a goto homework to eat even more of the already limited students' time with busywork.
I think it's probably more concerning that spitting out the most generic mathematically formulaic bullshit on a subject is likely to get a decent mark. In that case what are we actually testing for?
Can't perform any research, compare conflicting sources, or self-reflection.
It was not just "turn in the paper at the end" but turn in your topic with a paragraph describing it. Then make an outline, then bibliography of the sources we were using. During the process we had to use 3x5 index cards with various points, arguments, facts, and the specific pages in the books listed in our bibliography. We did this because this was later used to make footnotes in our paper.
By structuring the project this way and having each milestone count as 5-10% of the overall grade it made it much harder to cheat and also taught us how to organize a research paper.
I suppose you could ask ChatGPT to do the entire paper and then work backwards picking out facts and making the outline etc.
There's a good question about the future and utility of long at-home research paper projects in school, but it's not a cornerstone of education.
In 9th grade I procrastinated the semestral paper so much that I bought an essay online that explored unexpected gay themes in Ray Bradbury's corpus of work. I was so lazy I didn't even read it first, only skimmed it, and then back to Runescape. So it's not like this is a new problem due to LLMs, and I think take-home semester projects are all quite bad for these reasons that predate LLMs.
(It turned out to be such a phenomenally audacious essay that my teacher started fascinated email correspondence with me about it and I was forced to not only study the essay but also read the quoted parts of his work. Ugh, backfire.)
One of the reasons I mentioned the viva was an example of how we can decouple production of some work from an assessment of quality and some reason to believe that some candidate is capable of the work without assistance.
It would be unreasonable to spend five or so years working under examination conditions. But that doesn’t mean we can’t subsequently examine a candidate to determine likely authorship amongst other things.
But, for example, high school AP English exam is 3 45 minute essays (plus multiple choice). You have the read the passages, compare/contrast, etc.
Same with homework. If everyone has to solve the same 10 problems, divide and conquer saves everyone a lot of time.
Of course, you're only screwing yourself because you'll negatively impact your learning, but that's not something you can easily convince kids of.
In person oral exams (once you get over the fear factor) work best, with or without (proctored!) prep time.
Maybe it doesn't scale as well, but education is important enough not to always require maximal efficiency.
This assumes that homework helps kids learn, or that the knowledge required to succeed in school will help kids once they graduate.
I still feel like AI detectors would work well if we have access to the exact model, output probabilities of tokens, We can just take a bit of given text, and calculate the cumulative probability that the AI would complete it exactly like that.
It never was. It's just even more ineffective now that AI exists, than before.
The central example of this is college admissions statements. Some kids have the advantage both of parents who can afford to give them the experiences that look good on such an essay (educational trips to Africa, lessons in two musical instruments, one-on-one golf coaching, that kind of thing), and who can hire tutors to "support" them in writing the essay. AI just makes the tutor part accessible/affordable for a wider segment of the population.
It would be naive to assume that, pre-AI, there was not a "gray" essay-coaching market as well as the "dark" essay-writing as a service market. That market still works better than AI in many cases.
This is an absolute disgrace. And then these are the people who lecture you on "inclusion".
Are they? Is there any evidence of correlation between these two groups of people?
I believe it's still true today, but with future AI systems even highly abstract math is under the danger.
The poem was assigned to us, but for some reason the subject matter really chimed with me personally. I thought about it a lot, and—as a result—ended up writing a great essay.
Because I did well, I was accused of cheating in front the class.
Teachers are definitely fallible.
Let's not assume a lot right now. OpenAI and other companies are torching through cash like drunken socialist sailors. Will AI be here as a Big Data 2.0 B2B technology? Most likely, but a viable model where students and laypeople have access to it? To be seen.
We all mooched off of dumb VC money at one point or another. I acquired a few expensive watches at Fab dot com at 80% off when they were giving money away, eh.
You can run GPT-4-equivalent models locally. Even if all software and hardware advancements immediately halt, models at the current level will remain available.
2040? 2050?
In isolation, I don't think a model necessarily becomes less useful over time. It'll still be as good at summarizing articles, translating text, correcting grammar, etc. for you as it is today.
If things do continue to advance and new models are released, which I think is likely, the old ones become less useful by comparison and in situations where there's competition. But then, through hardware/algorithmic improvements, better models also become feasible for universities/open-source groups/individuals - so you shouldn't be stuck with a 2024 era model.
Sorry, English is not my first langage what is this expression ? Why does the sailor as to be socialist ? Google didn't help me with this one.
This is similar to how the AI companies mostly spending VC’s money buying these accelerators from nVidia.
If done right they would go from VC money with an expected exit to government money that overpays for incompetence because our only way out of deficit spending is through more debt and inflation.
The process of reading textual material, thinking about it, and then producing more textual material about the stuff you just read (and maybe connecting it to other stuff that you've read in the past) is a critical way of developing thinking skills and refining your ability to communicate to an audience.
The value of that shouldn't be overlooked just like the value of basic numeracy shouldn't be overlooked because we all carry calculators.
You're right that it would be better if post secondary institutions would test people's ability to think in more ways than just what they can regurgitate onto a piece of paper, if only because that can be easily cheated but that doesn't mean that there isn't personal benefit in the experience of writing an essay.
I may not be the best writer but I am a better writer because I wrote essays in university, and I may not be great at math but I can reason and estimate about a variety of things because I have taken many math courses. These things have ultimately made me a better thinker and I am grateful to have had that imparted to me.
However, +90% of students will not now do any of that work. I got out of teaching (coincidentally) before LLMs appeared, and even then +80% of students did not experience that benefit of the essay process even with a grade (and plagiarism consequences) to motivate them. Now that decent-ish prose is a few keystrokes or Siri-led "chats" away, that's what they're going to do. That's what they're going to do.
I know of - I think it's up to four, now - former colleagues taking early retirement, or changing careers, rather than continue teaching Humanities in a world of LLMs.
So the obvious solution is to make students to talk with an AI, which would grade their performance. Or, maybe the grading itself could be done by a minimum wage assistant, while AI would lead the discussion with a student.
It is, because I'm becoming tired with the current AI hype. It lasts too long to be funny.
OTOH, professor talking with a student is a good way to assess the academic performance of the student, but there are some caveats beyond costs. For example, professor will struggle to be an objective judge. Moreover even if they succeed, they would face accusations of discrimination in any case.
AI could solve this problem, but I'm not sure if AIs will be up to a task of leading the discussion. Though maybe if you try to assess students on their ability to catch AI on a hallucinated bullshit...
Why not have the testing done externally, by really neutral persons?
But AIs and especially LLMs are way too unreliable for the foreseeable future.
It seems to me that this is pretty much immune to plagiarism as well as being much better for the student.
When you audition for a symphony you perform behind a curtain and are required to wear soft slippers (so they can't tell if you are a wearing high heals - female).
We can probably use voice changers so the examiner cannot tell who you are by your voice, but those tend to be fatiguing.
Not saying this is a fatal flaw, but there is a bit of a tradeoff there.
Of course the current setup works badly for those who explain it much better by speaking.
Those of you who severely lack any of the three will be penalized. Just like someone who can discuss a topic orally but could not write it up would be penalized.
Yes, there are people who have trouble with public speaking to a debilitating degree, but it would be excessively rare for someone to not at the very least in a one on one with their professor/teacher be able to be so badly affected as to not seem they've even heard of a topic or at least be able to prove they've worked on it to a certain degree.
I would be immediately skeptical of any student who claims they are completely unable to explain their knowledge unless they are allowed to work in complete isolation with nobody to monitor they aren't cheating in some way.
Lots of elite prep schools in the US use a similar system, for similar reasons.
I think the hard answer is that to some extent you just have to learn to. I mean, you could sit silently in supervisions if you really insisted, but to participate properly you just needed to build the confidence.
Is it fun? No, but it's a pretty accurate reflection of life after school: nobody in the real world gives you points for "couldn't say the right thing at the right time, but was thinking it"
Can you communicate it in real-time through writing? Maybe that's an accommodation that could be done?
Yeah, the General Problems exam was a nightmare, I think the professors competed each year to come up with the toughest questions. Getting 50% was an excellent score.
It did force you to learn all the material though, especially as at the end of 3-4 years you may have forgotten some of it, like Optics or whatever. It was pretty hardcore though, especially compared to my friends studying other subjects.
However, you also mentioned a final project. You’d be shocked how much commissioning exists where people have their projects produced for them. I’m not talking an overly helpful study group, I mean straight up essay mills. Tools like ChatGPT make the bar for commissioning lower and cheaper. I don’t know how you can combat this and still have long-term projects like dissertations.
Many years later I took a course in C# at a university in Norway and that was not merely open note but also open book (you could take the set book in). Again that gives the exam author the possibility to really discover who knows what.
I doubt that your rich middle-western student would have passed either of these
I doubt that anyone could have produced a plausible report without actually doing the work. And to defend it one would have to understand the underlying physics and the work that was done. Plus I think my supervisor and the other two students who worked with me on the project would have remarked on my absence from the laboratory if I had simply bought the paper!
You can still have long term projects and dissertations so long as the degree is awarded for the defence of the dissertation rather than the dissertation itself; that is the student must demonstrate in a viva that they understand everything in the dissertation rather than merely regurgitate it.
There are certainly similar projects being completed by students every year, and doubtless those students are not cheaters, but for each dissertation like yours, there are probably 10 or more projects that are not collaborative and have no artefacts or supporting evidence other than a written report. Such projects are fairly easy to commission. For a reasonable price (potentially thousands of dollars) you can pay a poor research student in the same field as you to churn out a mid-tier dissertation. This can be detected with a viva, but the academics need to be very confident before accusing someone of cheating. More often than not, you can get away with it and just get a not great grade.
I think that in general the natural/formal sciences don't suffer nearly as much as social science and humanities do, simply because exams and labs tend to highlight irregularities, and cheaters are less likely to be drawn into "hard" fields. However, it still exists in every field.
Similar re: exams, they were available but sticking rigidly to them didn’t help much.
For longer essays, you can just build in an oral examination component. This face time requirement is just not that hard to include given that even in lecture hall style settings you can rely on graduate student TAs who do not really cost anything. The thing is that the universities don't want to change how they run things. Adjuncts in most subjects don't cost anything and graduate students don't cost anything. They earn less than e.g. backroom stocking workers. This is also why they, by and large, all perform so poorly. 30 minutes of examiner time costs maybe $11 or less. Even for a lecture class with 130 students, that's under $1,500. Big woop.
There are some small changes to grading practices that would make life very hard for AI cheaters, such as even cite checking a portion of citations in an essay. The real problem is that US universities are Soviet-style institutions in which gargantuan amounts of cash are dumped upon them and they pretend to work for it while paying the actual instructors nothing.
Hiring a n TA to delegate grading that’s hard to verify seems like will cost more than you think.
In fact, the only reason I use the word 'importunate' to describe myself, is because that's what my undergrad advisor called me.
So I at least was able to get well over 30m with each professor to discuss whatever I wanted. But likely that's b/c there wasn't a lot of competition.
> Written assay evaluation is not and has never been an effective evaluation.
I've kept a blog for almost 20 years now and one thing is for sure: well-structured writing is very different from an oral exam the writing allows for restructuring your thoughts and ideas as you go and allows for far more depth.
I don't think, for most folks, that they could have as much depth in an F2F as they could in their writing with the exception of true experts in their fields.
The written essay has a cohesiveness and a structure to it that provides a better framework for evaluation and conveyance of information.
I'm not so sure that writing takes more time than studying. For starters, you don't have to memorize anything, and you can limit yourself to the assigned topic.
Of course, it can be that students don't take studying for an oral exam seriously, and trust the teacher to only ask superficial questions.
That strikes me as a workable bottom line.
Average student debt after a 4 year degree is ~$35,000 after ~45 courses. Before even running the math it should be obvious the gigantic cost of higher ed over 4 years is entirely unrelated to what an instructor would be making for ~23 hours of work (barring a secret society of multi millionaires). I.e. the problem you're identifying is the vast majority of $ spent in higher ed is not going to time with your professors, not that doing so is itself expensive.
Could not disagree more. Researching, formulating arguments, can give a student a complete view of the subject that studying for tests misses. But, similarly to tests, it probably depends on the skill of the teacher in creating the right kind of written assignments.
On a more serious note - US removed cursive from their curriculum almost two decades ago - something i cant wrap my head around as cursive is something the rest of the world(?) uses starting in middle school and onwards through the whole adult life.
There are lots of things I spent a lot of time learning school that I rarely use, but see the value in having learnt. Cursive, beyond a very basic level, is not one of those things.
Though I’m no education expert, perhaps there is a subliminal value to spending all that time.
I learned cursive, then reverted to print, but when I entered a phase of my life where I needed to write several pages a day I quickly went back to (a custom variant of) cursive because it was faster to write in a legible way than print.
When I rush print it quickly becomes illegible. When I rush my cursive it doesn't look quite as nice as it does when I'm writing steadily, but I can still read what I wrote ten years later.
From what I can tell it works because cursive letters are defined in a shape that lends itself to a quick moving pen. Once you learn that shape (both to write and read), you can quickly get words down on a page and then understand them later. If you just try to slur your print in an unprincipled way your letters distort in ways that make them harder to tell apart.
Now, I imagine someone could develop a slurred print that doesn't have connections between letters, but I'd probably call that a cursive anyway.
Later I found out it was developed for use with a fountain pen, designed with the idea that a correctly-faced nib would make some strokes bolder and others very faint, and to keep the nib always moving in a kind of flow to avoid spots, and to make it natural to keep the nib faced the correct way(s), plus with even more attention to avoiding raising the pen than most cursives, for similar reasons of avoiding spotting. That made all the downsides make sense—it's far less ugly and easier to read when written with a fountain pen, and may well be faster than many other similarly-clean methods of writing with one.
Why the hell we were still learning that hand decades into the dominance of the ballpoint, remains a question.
I did nearly all my exams on a computer.
At one point the best writing tool was the fountain pen. It was a great invention and it had an appropriate script: cursive, which was the natural thing to do given how the ink flowed.
However kids are messy and you really want them to use pencils because they don't have flowing ink. The reason for cursive in the first place was the flowing ink, so when we switched away from flowing ink, there was no reason to write in cursive.
Except of course to waste the only resource everybody agrees is okay to waste: kids time.
The time would be far better spent on say personal finance lessons, which people seem to really struggle with. What's a mortgage, what's a loan, what gets taxed, that sort of thing.
What happens if you simply print your statement like most people who need to write legibly do for the majority of their life?
At the end of the day cursive writing is a hobby, not a skill. We don't need it anymore. It wastes priceless learning time at a critical juncture in our intellectual development.
Here is the brand new method - asking verbal questions in person and evaluating answers. Also allow high tech aides in the form of chalk and blackboard
For all their problems, 5000 word take home assignments in Microsoft Office have a lot in common with the activities of a junior management consultant, NGO writer, lawyer or business analyst. And same with for scientists but with Latex.
I’d rather hire a lawyer who could only do their job with AI than one who couldn’t use a computer to create documents or use digital tools to search case law.
You are in a room with a sheet of paper and a pen. Go.
You’re acting as if 2010 was a hundred years ago.
Calculators are just one analogy, there is no guarantee it will work out that way. It's just as likely that this over-technologization of the classroom will go the way of whole-language reading education.
While writing term papers is a skill that is only minimally useful in the real world (save for grant writers and post docs, pretty much), the patterns of thinking it encourages are valuable to everything that isn’t ditch digging.
Maybe we can outsource this part of our cognition to AI, but I’m skeptical of the wisdom of doing so. Are we all going to break to consult ChatGPT in strategy meetings?
That was just a simple quip to shut down student bellyaching. Even before we had pocket calculators, it was never a strong answer. It just had to hold over long enough so when you realized it was bad answer you weren't that teacher's problem anymore.
The actual answer was that they're complaining about a minor inconvenience designed for reinforcement, and if they really did need a calculator for the arithmetic on a test designed deliberately designed to be taken without a calculator, then they don't belong in that class.
Now this would effectively kill off the current AI powered solution, because they have no way of explaining, or even understanding, why a paper may be plagiarized or not, but I'm okay with that.
This is a big part of GDPR.
> The data subject shall have the right not to be subject to a decision based solely on automated processing [...]
[1]: https://gdpr.eu/article-22-automated-individual-decision-mak...
But if you're a nobody, and can't afford to go to court against Deutsche Bank for example, of course you're SOL. EU has some good parts, but it's still a human government.
It's especially problematic since a good chunk of those "flagged" are actually doing something nefarious, and both courts and government will consider that "mostly works" is a good outcome. One or ten unlucky citizens are just the way the world works, as long as it's not someone with money or power or fame.
The assumption is that a human review the conditions that led the automated system to make that decision.
I think it would be trivial to argue in court that rubberstamping some scalar value that a deep neural net or whatever spit out does not pass that bar. It's still the automated system's decision, the human is just parroting it.
Note that it's easier for the FAANGs to argue such a review has happened because they have massive amounts of heterogenous data where there's bound to be something that would be sufficient to argue with (like having posted something that offended someone).
But a single score? I'd say almost impossible to argue. One would have to demonstrate that the system is near-perfect, and virtually never makes mistakes.
Reading the rules quickly, it does seem like you're not entitled to know why the computer flagged you, only that you have the right to "obtain human intervention". That seems a little to soft, I'd like to know under which rules exactly I'm being judged.
I also thing that there should be laws requiring a clear explanation whenever that happens.
The problem being discussed here feels like it should be similar in that last regard: any time an automated system is making a serious decision they should be required to have an explanation and review process. If they don’t have sufficient evidence to back up the claim, they need to collect that evidence before making further accusations.
That's not correct. Some solution look at perplexity for specific models, some will look at ngram frequencies, and similar approaches. Almost all of those can produce a heatmap of "what looks suspicious". I wouldn't expect any of the detection systems to be like black boxes relying on LLM over the whole text.
In any case, if you where to use LLMs, or other black box solutions, you'd have to yank those out, if you where met with a requirement to explain why something is suspicious.
Anyone interested to learn more about it, I recommend the recent book "AI Snake Oil" from Arvind Narayanan and Sayash Kapoor [1]. It is a critical but nuanced book and helps to see the whole AI hype a little more clearly.
[1] https://press.princeton.edu/books/hardcover/9780691249131/ai....
Reliable systems in some areas? - Absolutely, and yes, even facial recognition. I agree, it works very well, but that is a different issue as it does not reveal or try to guess anything about the inner person. There are other problems that arise from the fact that it works so well (surveillance, etc.), but I did not mean that part of the equation.
Either there can be an undefeatable AI detector, or an undetectable AI writer, both can't exist in the same universe. And my assumption is that with sufficient advances there could be a fully human-equivalent AI that is not distinguishable from a human in any way, so in that sense being able to detect it will actually never work.
Examples: Spam detection, copyrighted material detection, etc.
Really? A spammer is trying to ace a test where my attention is the prize. I don't really see a huge difference between a student/diploma and a spammer/my attention.
Education tech companies have been playing with ML and similar tech that is "AI adjacent" for decades. If you went to school in the US any time after computers entered the class room, you probably had some exposure to a machine generated/scored test. That data was used to tailor lessons to pupil interest/goals/state curricula. Good software also gave instructor feedback about where each student/cohort is struggling or not.
LLMs are just an evolution of tech that's been pretty well integrated into academic life for a while now. Was anything in academia prepared for this evolution? No. But banning it outright isn't going to work
You don't see a difference between potentially ruining a students future due to grading done by an opaque ai system and you clicking on a spam email? That's preposterous.
I had both, over a decade ago in high school. Plagiarism detection is the original AI detection, although they usually told you specifically what you were accused of stealing from. A computer-based English course I took over the summer used automated grading to decide if what you wrote was good enough (IIRC they did have a human look over it at some point).
This happens anyways, though? Any service that's useful for alternative / shady / illicit purposes is part of a cat/mouse game. Even if you don't tell the $badActors what you're looking for, they'll learn soon enough what you're not looking for just by virtue of their exploitative behavior still working.
I'm a little skeptical of any "we fight bad guys!" effort that can be completely tanked by telling the bad guys how they got caught.
In what world is this fair? Our court systems certainly don't operate under these assumptions.
Figuring out who the hell you are in your high school years was hard enough when Kafka was only a reading assignment.
GDPR to the rescue!
https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-re...
You must identify whether any of your processing falls under Article 22 [automated decision making, including AI] and, if so, make sure that you:
* give individuals information about the processing;
* introduce simple ways for them to request human intervention or challenge a decision;
* carry out regular checks to make sure that your systems are working as intended.
Why in gods name has the USA not adopted similar common sense legislation?
So, now I use ChatGPT to check my English. I just write what I want to write than ask it to make my text more "More concise, business-like and not so American" (yeah the thing is by default as ultra enthusiastic as an American waiter). And 9 out of 10 times it says what I want to say but better than I wrote myself, and in much less words and better English.
I don't think it took less time to write my report, but it is much much better than I could have made alone.
AI detector may go off (or it goes on? of is it of? Idk, perhaps I should ask Chat ;)), but it is about as useful as a spell-check detector.
It's a Large Language Model, you should just is like that, it is not a Large Fact Model. But if you're a teacher you should be a good bullshit detector, right?
If I'm every checking some student's report, you may get this feedback: For god's sake, check the language with ChatGPT, but for God's sake check the fact in some other way.
Making texts more accessible through clear language and well-structured arguments is a valuable service to the reader, and I applaud anyone who leverages LLMs to achieve that. I do the same myself.
With students, historically we have always assumed that written communication was the more challenging skill and our tests were arranged thusly. But we're in a new place now where the inability to verbally converse is a real hurdle to overcome. Maybe we should rethink how we teach and test.
Both classes got a lesson, from either end, essentially for free (for the teacher). And it really helped. The next year I got to do the same. Of note was that this was back in the day when computers were relatively rare and typing was a skill that was specially taught, so most of the papers were written longhand for the first draft.
It's long been said that if you really want to learn a subject you should teach it. This sort of give-and-take works well, and it is more or less how the rest of society works. Using AI for this would be quite similar, but I think having another human is better. An AI will never stop you in the hall and say "dude, your paper, I got totally lost in the middle section, what the hell," but sometimes that's quite helpful.
Not by design, but the training corpus necessarily includes a lot of "facts" (claims made by whoever wrote the original text). A model that is trying to output nonfiction on a specific topic, is likely to encounter relatively more models of claims that either actually were incidentally true, or at least have the same general form as true claims without an obvious "tell".
Of course, every now and then it goes off the rails and "hallucinates". Bad luck for the student who doesn't verify the output when this happens (which is probably a lot of students, since part of the motivation to cheat is not knowing the material well enough to do such verification properly).
However, the AI identifies the school issued Lenovo laptops as weapons. So every kid was flagged. Rather than stopping using such a stupid tool, they just have the kids remove their laptops before going through the scanner.
I expect not smart enough people are buying “AI” products and trusting them to do the things they want them to do, but don’t work.
But if they are not otherwise checked it would be quite useless.
The problem is, if it has been that badly tested that it detects Lenovo laptops as weapons, there is a good chance that it doesn't properly detect actual weapons either.
There hasn't been a school shooting in the UK for nearly 30 years. Handguns were banned after the last school shooting and there hasn't been one since.
https://en.wikipedia.org/wiki/Category:School_shootings_in_t...
Although that fact is sometimes forgotten by schools who copy the US in having "active shooter drills" though. Modern schools sound utterly miserable.
got a license for that mate?
The US has more stabbings per-capita than the UK does, even on top of the shootings.
Licensing wouldn't be worthwhile as almost every household would want knives for food preparation.
It would indeed be a miracle as the US would drastically have to reduce their number of shootings to get down to the number of UK mass stabbings.
https://en.wikipedia.org/wiki/List_of_mass_stabbings_in_the_...
https://en.wikipedia.org/wiki/List_of_mass_shootings_in_the_...
It seems to me like victim blaming for U.S. schools to have active shooter drills - it makes more sense to have much better training and screening of gun owners than trying to train the victims. However, given that the NRA is excessively powerful in U.S. politics, I can see why they are necessary, but it just seems easier to me to stop kids from being able to get hold of guns (e.g. have some rudimentary screening for gun purchases and require owners to keep them in locked cabinets when they are not in use).
[1] https://en.wikipedia.org/wiki/List_of_countries_by_intention...
People are willing to believe almost anything as long as it makes their lives a little more convenient.
I want to emphasize, that this isn't really about trusting magic, it's about people nonchalantly doing ridiculous stuff nowdays and that they aren't held accountable for that, apparently. For example, there were times back at school when I was "accused" of cheating, because it was the only time when I liked the homework at some class and took it seriously, and it was kinda insulting to hear that there's absolutely no way I did it, but I still got my mark, because it doesn't matter what she thinks if she cannot prove it, so please just sign it and fuck off, it's the last time I'm doing my homework at your class anyway.
On the contrary, if this article to be believed, these teachers don't have to prove anything, the fact that a coin flipped heads is considered enough of a proof. And everyone supposedly treats it as if it's ok. "Well, they have this system at school, what can we do!" It's crazy.
That's how you can mold society as you like at your level: this student's older sibling was a menace? Let's fuck them over, being shitty must run in the family. You don't like the race / gender / sexuality of a student? Now "chatGPT" can give you an easy way to make their school life harder.
Just introduce an incomprehensible process, Like applying for a Visa or planning permission, and then use it to your advantage.
From the victims perspective, there is no difference between bureaucracy and AI
I agree. But now some people can point to ChatGPT or other tools and use it as an excuse. So for them, the "bugs" are a feature. They don't care about false positives, they care about the fact some authority tells them a student they don't like used AI to write an essay.
People trust a system because other people trust a system.
It does not matter if the system is the inquisition looking for witches, machine or Gulag from USSR.
The system said you are guilty. The system can’t be wrong.
Kafka is rolling in his grave.
There was recent article about yet another generated text in the US court, this time without malicious intent (it seems). The article boils down to the fact that the plaintiff asked neural network to do a historical financial calculation of property cost and immediately trusted it, "because computers". Computers are always correct, NNs run on computers, hence they are always correct :) . Soon this mentality will be in every household on the planet. We will be remembering days of media dishonesty and propaganda with fondness, at least previously we kinda could discern if the source was intentionally lying.
Well, you shouldn’t be so surprised. You just described 95%+ of the population’s approach to any form of technology. And there’s very rarely any discomfort with such ignorance, nor any desire to learn even the basics. It’s very hard to understand for me — some of us just have to know!
I’ll give you a hint: they’re not ignorant.
We don't do calculations: computers do it for us.
We don't accumulate knowledge: we trust Google to give us the information when needed.
Everything in a small package everyone can wear all day long. We're at the second step of transhumanism.
I got AI answer saying ‘no’, but actually you do.
If I use a calculator it will be correct. If I open encyclopaedia it will mostly be correct, because someone with a brain did at least 5 minutes of thining.
We are not talking about some minor detail, AI makes colossal errors with great confidence and conviction.
GPS is great at knowing where you are, but directions are much much harder, and the extra difficulty is why the first version of Apple Maps was widely ridiculed.
Even now, I find it's a mistake to just assume Google Maps can direct me around Berlin public transport better than my own local knowledge — sometimes it can, sometimes it can't.
(But yes, a single original Pi Zero beats all humans combined at arithmetic even if all of us were at the level of the world record holder).
After I moved here and learned the system, I realised it had on my first trip directed me through a series of unnecessary train routes for a 5 minute walk.
Last summer, when trying to find a specific named cafe a friend was at, Google Maps tried to have me walk 5 minutes to the train station behind me to catch the train to the stop in front of me to walk back to… the other side of the street because I hadn't recognised the sign.
It's a great tool, fantastic even, but it still doesn't beat local knowledge. And very occasionally, invisibly unless you hit the edge, the map isn't correctly joined at the nodes and you can spot the mistake even as a first time visitor.
To be honest, we do for most things: I have not checked the speed of light. And I surely would not be able to implement a way to measure it from only my observations and experience.
At least not having to care about arithmetic leaves more time to care about mathematics.
I mean I'm skeptical about AI as well and don't like it, but I can see it becoming a force multiplier itself.
Posters here love to bring out this argument, but I think a major weakness is that those people wound up being right. People don't memorize things any more! I don't think it's fair to hold out as an example of fears which didn't come to pass, as they very much did come to pass.
....
and it made no difference.
The easiest for someone here to see is probably code generation. You can point at parts of it and go "this part is from a high-school level tutorial", "this looks like it was grabbed from college assignments", and "this is following 'clean code' rules in silly places"(like assuming a vector might need to be Nd, instead of just 3D).
If they could make it elsewhere, they would.
I don’t expect this to be a popular take here, and most replies will be NAXALT fallacies, but in aggregate it’s the truth. Sorry, your retired CEO physics teacher who you loved was not a representative sample.
Switzerland and Finland comes to mind.
https://www.swissinfo.ch/eng/society/swiss-salaries-teachers...
110k in Switzerland is a good pay today. The article is from 2017.
Hey, he was Microsoft’s patent attorney who retired to teach calculus!
"and then a bunch" is somewhat misleading. They in fact take easier and fewer classes in the subjects that they are studying for, but they have to take extra classes on education, which afaik are not that hard to pass. Getting a "Lehramt" degree is much easier than getting the regular degree in a subject, which is why many people that are simply not good enough for the real thing do it.
Also we have a teacher shortage and more and more teachers are not in fact people that received an education you usually have to get as a teacher, but are just regular people with either a degree in the subject they are teaching or a degree in almost anything (depends on how desperate the schools are and what subjects they are hiring for).
Because if you're putting forth the assertion "If they could make it elsewhere, they would." you've certainly had spent sometime teaching, yes?
I think it would be good to understand how much experience teaching it took for you to come to that conclusion.
Anyone who's been around AI generated content for more than five minutes can tell you what's legitimate and what isn't.
For example this: https://www.maersk.com/logistics-explained/transportation-an... is obviously an AI article.
to some degree of accuracy.
A. The Pathfinder and The Deerslayer stand at the head of Cooper's novels as artistic creations. There are others of his works which contain parts as perfect as are to be found in these, and scenes even more thrilling. Not one can be compared with either of them as a finished whole. The defects in both of these tales are comparatively slight. They were pure works of art.
B. The Pathfinder and The Deerslayer stand at the head of Cooper's novels as artistic creations. There are others of his works which contain parts as perfect as are to be found in these, and scenes even more thrilling. Not one can be compared with either of them as a finished whole. The defects in both of these tales are comparatively slight. They were pure works of art.
https://www.gutenberg.org/files/3172/3172-h/3172-h.htm#:~:te....
I think it is, however the dream among educators of an “AI detector” is so strong that they’re willing to believe “these guys are the ones that cracked the problem” over and over, when it’s not entirely true. They try it out themselves with some simple attempts and find that it mostly works and conclude the company’s claims are true. The problem though is that their tests are all trying to pass off AI-generated work as human-generated—not the other way around. Since these tools have a non-zero false positive rate, there will always exist some poor kid who slaved away on a 20-page term paper for weeks that gets popped for using AI. That kid has no recourse, no appeals—the school spent a lot of money on the AI detector, and you better believe that it’s right.
This is what will eventually happen. Some component or provider deep in the stack will provide some answer and organizations will be sufficiently shrouded from hard decisions and be able to easily point to "the system."
This happens all the time in the US. Addresses are changed randomly because some address verification system feedback was accepted w/o account owner approval -- call customer service and they say "the system said that your address isnt right", as if the system knows where i've been living for the past 5yrs better than me, better than the DMV, better than the deed on my house. If the error rate is low enough, people just accept it in the US.
Then, it gets worse. Perhaps the error rate isnt low, just that it is high for a sub-group. Then you get to see how you rank in society. Ask brown people in 2003-2006 how fun it was to fly. If you have the wrong last name and zipcode combo in NYC suddenly you arent allowed to rent citibikes despite it operating on public land.
The same will happen with this, unless there is some massive ACLU lawsuit which exposes and the damages will continue until there is a resolution. Quite possibly subtle features on language style will get used as triggers, probably unknowingly. People in the "in-group" who arent exposed will claim it is a fair system while others will be forced to defend themselves and have to provide the burden of proof on a blackbox.
I suspect there is a product opportunity here. It could be as simple as a chrome extension that records your sessions in google docs and generates a timelapse of your writing process. That’s the kind of thing that’s hard to fake and could convince an accuser that you really did write the essay. At the very least it could be useful insurance in case you’re accused.
what.
"One of my kid's teachers set out a warning to students that all essays would be checked against the other students' essays to see if they are the same and the repercussions one would face if caught. A classmate did a Google search and found the questions of the essay as examples on a book."
One thing is perfectly valid, the other one is not.
Then of course, there are shades of gray. Using ChatGPT for some things is not copying and you can even say the kids are learning to use the tool, but if you use it for 95% of the essay, it is.
This isn't a made up situation. Teachers at my school have used AI for essay prompts, test questions, etc and it spreads around and generally leads to the sentiment that "if the teacher is doing it, they can't in good faith tell me to not". Imagine if in math class the teacher , after just telling the students they can't use a calculator, types in a simple arithmetic expression into their calculator.
Doubtful. This is a new sector/era in the cat-v-mouse game.
> we learn to adapt away from graded homework.
Nothing proposed as an alternative scales well and - ironically - it's likely that something _like_ an LLM will be used to evaluate pupil quality / progress over time.
I'm asking, because all this "AI" text-generation stuff isn't a technology problem. It's 101% a human problem.
However programming a computer to do this isn't easy. In a previous job I had dealing with plagiarism detectors and soon realised how garbage they were (and also how easily fooled they are - but that is another story). The staff soon realised what garbage these tools are so if a student accused of plagiarism decided to argue back then the accusation would be quietly dropped.
//
And no, those were not generally followed by a real comment.
I've seen exactly what you describe and worse *, and I've also seen them keep to one style until I got bored of prompting for new features to add to the project.
* one standard test I have is "make a tetris game as a single page web app", and one model started wrong and then suddenly flipped from Tetris in html/js to ML in python.
Easy to catch people that aren't trying in the slightest not to get caught, right? I could instead feed a corpus of my own writing to ChatGPT and ask it to write in my style.
I assume most people would make least amount of effort and simply prompt chat interface to produce some text, such text is rather detectable. I would like to see some experiments even for this type of detection though.
The point of these detectors is to spot stuff the students didn’t research and write themselves. But if the corpus is your own written material then you’ve already done the work yourself.
To get an LLM to generate content in your own writing, there’s going to be no substitute for training it on your own corpus. By which point you might as well do the work yourself.
The whole point cheating is to avoid doing the work. Building your own corpus requires doing that work.
You're saying it's not good enough at mimicking styles. others saying it's good enough. I think if it's not good enough today it'll be good enough tomorrow. Are you betting on it not becoming good enough?
Teachers will notice if students writing style shifts in one piece compared to another.
Nobody disputes that you can get LLMs to mimic other people. However it cannot mimic a specific style it hasn’t been trained on. And very few people who are going to cheat are going to take the time to train an LLM on their writing style since the entire point of plagiarism is to avoid doing work.
Imagine the teacher saying "this is not your style it's too good" to a student who legit tried killing any motivation to do anything but cheat for remaining life
If the student always uses LLMs then it would be pretty obvious by the fact that they’re failing at the cause in all bar the written assessments (ie the stuff they can cheat on).
> Also do you expect that student's style is fixed forever
Of course not. But people’s styles don’t change dramatically on one paper and reset back afterwards.
> teachers are all so invested that they can really tell when the student is trying something new vs use an LLM that was trained to output writing in the style of an average student?
Depends on the size of the classes. When I was at college I do know that teachers did check for changes in writing styles. I know this because one of the kids on my class was questioned about his changes in his writing style.
With time, I’m sure anti-cheat software will also check again previous works by the students to check for changes in style.
However this was never my point. My point was that cheaters wouldn’t bother training on their own corpus. You keep pushing the conversation away from that.
> Imagine the teacher saying "this is not your style it's too good" to a student who legit tried killing any motivation to do anything but cheat for remaining life
That’s how literally no good teacher would ever approach the subject. Instead they’d talk about how good the paper was and ask about where the inspiration came from.
There's nothing stopping students from generating an essay and going over it.
>Of course not. But people’s styles don’t change dramatically on one paper and reset back afterwards.
Takes just a little effort to avoid this.
>With time, I’m sure anti-cheat software will also check again previous works by the students to check for changes in style.
That's never going to happen. Probably because it doesn't make any sense. What's a change in writing style ? Who's measuring that ? And why is that an indicator of cheating ?
>However this was never my point. My point was that cheaters wouldn’t bother training on their own corpus. You keep pushing the conversation away from that.
Training is not necessary in any technical sense. A decent sample of your writing in the context is more than good enough. Probably most cheaters wouldn't bother but some certainly would.
This then comes back to my original point. If they learn the content and rewrite the output, is it really plagiarism?
> Takes just a little effort to avoid this.
That depends entirely on the size of the coursework.
> That's never going to happen. Probably because it doesn't make any sense. What's a change in writing style ? Who's measuring that ? And why is that an indicator of cheating ?
This entire article and all the conversations that followed are about using writing styles to spot plagiarism. It’s not a new concept nor a claim I made up.
So if you don’t agree with this premise then it’s a little late in the thread to be raising that disagreement.
> Training is not necessary in any technical sense. A decent sample of your writing in the context is more than good enough. Probably most cheaters wouldn't bother but some certainly would.
I think you’d need a larger corpus than the average cheater would be bothered to do. But I will admit I could be waaay off in my estimations of this.
Who said anything about rewriting? That's not necessary. You can have GPT write your essay and all you do is study it afterwards, maybe ask questions etc. You've saved hours of time and yes that would still be cheating and plagiarism by most.
>This entire article and all the conversations that followed are about using writing styles to spot plagiarism. It’s not a new concept nor a claim I made up.
>So if you don’t agree with this premise then it’s a little late in the thread to be raising that disagreement.
The article is about piping essays into black box neural networks that you can at best hypothesize is looking for similarities between the presented writing and some nebulous "AI" style. It's not comparing styles between your past works and telling you just cheated because of some deviation. That's never going to happen.
>I think you’d need a larger corpus than the average cheater would be bothered to do. But I will admit I could be waaay off in my estimations of this.
An essay or two in the context window is fine. I think you underestimate just what SOTA LLMs are capable of.
You don't even need to bother with any of that if all you want is a consistent style. A style prompt with a few instructions to deviate from GPT's default writing style is sufficient.
My point is that it's not this huge effort to have generated writing that doesn't yo-yo in writing style between essays.
Maybe. But I think we are getting too deep into hypotheticals about stuff that wasn’t even related to my original point.
> The article is about piping essays into black box neural networks that you can at best hypothesize is looking for similarities between the presented writing and some nebulous "AI" style. It's not comparing styles between your past works and telling you just cheated because of some deviation. That's never going to happen.
You cannot postulate your own hypothetical scenarios and deny other people the same privilege. That’s just not an honest way to debate.
> My point is that it's not this huge effort to have generated writing that doesn't yo-yo in writing style between essays.
I get your point. It’s just your point requires a bunch of assumptions and hypotheticals to work.
In theory you’re right. But, and at risk of continually harping on about my original point, I think the effort involved in doing it well would be beyond the effort required for the average person looking to cheat.
And that’s the real crux of it. Not whether something can be done, because hypothetically speaking anything is possible in AI with sufficient time, money and effort. But that doesn’t mean it’s actually going to happen.
But since this entire argument is a hypothetical, it’s probably better we agree to disagree.
performing badly under pressure is not a thing in your world
> My point was that cheaters wouldn’t bother training on their own corpus. You keep pushing the conversation away from that.
My point was cheaters don't need to train on their corpus. That's why it's zero effort. You keep trying to wave that away
> That’s how literally no good teacher would ever approach the subject.
Now we only need to eliminate bad teachers
No need to be rude.
Pressure presents different characteristics. Plus lecturers would be working with failing students so would understand the difference between pressure and cheating.
> My point was cheaters don't need to train on their corpus. That's why it's zero effort. You keep trying to wave that away
My entire point was that most cheats wouldn't bother training their corpus!
With the greatest of respect, have you actually read my comments?
> Now we only need to eliminate bad teachers
Well that's a whole other discussion :)
Good, because they don't need a custom corpus to cheat with LLMs with most normal teachers.
And if a teacher reduced your grade saying you are using LLM because your style doesn't match you just report them for it and say you were trying a new style (teacher would probably will be wrong 50% of the time anyway)
I think you're underestimating the capabilities of normal teachers. And I say this as someone who a large percentage of their family are teachers.
Also this topic was about using LLMs to spot LLMs. Not teachers spotting LLMs.
> And if a teacher reduced your grade saying you are using LLM because your style doesn't match you just report them for it and say you were trying a new style (teacher would probably will be wrong 50% of the time anyway)
You're drifting off topic again. I'm not going to discuss handling false positives because that's going to come down the policies of each institution.
Student don't have to perform research or study for the given task, they need to acquire an example of text suitable for reproducing their style, text structure, to create an impression of being produced by hand, so the original task could be avoided. You have to have at least one corpus of your own work for this to work, or an adequate substitute. And you still could reject works by their content, but we are specifically talking about llm smell.
I was talking about the task of detecting llm generated text which is incredibly hard if any effort is made, while some people have an impression that it's trivially easy. It leads to unfair outcomes while giving false confidence to e.g. teachers that llms are adequately accounted for.
I forgot the name of the guy, who said it, but he was some big philosophy lecturer at Harvard and his view on the matter ( heavy reading course and one student left a course review - "not reading assigned reading did not hurt me at all") was ( paraphrased):
"This guy is an idiot if he thinks the point of paying $60k a semester of parents money is to sit here and learn nothing.'
I could go on like this, but I won't. Each of us has a choice how we play the cards we are dealt.
I accept your point, but this point reinforces a perspective I heard from my accountant family member, who clearly can identify price, but has a hard time not equating it with value. I hesitate to use the word wrong, because it is pragmatic, but it is also rather wasteful ( if not outright dumb ).
Your characterization of an adult child does not seem fair. What makes someone an adult? If it's academic discourse, then why is it valuable?
I mean, if you're into it, more power to you. Go nuts with finally figuring out what makes a human. Just don't claim it's more virtuous than anyone else's hobby, unless you can find a reason.
To call it "wasteful" says that something of "value" is being squandered, but the value is perceived by each of us differently.
Hmm, I could try repeating the value argument from the other end, but lets approach it differently. You mention a credential itself being a value, which is not an unreasonable position to take, which is part of the reason I am not dismissing it outright.
But what is a credential? Oracle that is Google defines it as "a qualification, achievement, personal quality, or aspect of a person's background, typically when used to indicate that they are suitable for something." Are they qualified simply, because they 'put in their time' at an institution of supposed higher learning? If so, that credential is not only wasteful, it is also worthless. It exists for now only because it is riding on the glory of its past.
At least previous poster's argument was more direct: it is not about learning at all. It is a social club, where kids of already successful people are sent. That network of kids of the already successful has its uses, but if it is a social club AND, as we established in previous paragraph, it is a learning institution in name only ( used only for credential ), then it is really just that: a social club, which I characterized in a flight of fancy as an adult daycare. I do stand by this phrasing, because the more I think about it, the more it fits.
<< Your characterization of an adult child does not seem fair. What makes someone an adult?
Hmm, good question, but I will leave it unanswered. I want to see how you respond to the previous point.
<< If it's academic discourse, then why is it valuable?
Academic discourse is not valuable. Frankly, at its core, nothing is inherently valuable, because any value is the value we ascribe to things. You might think that this is me saying:
'ok, so anyone can value anything and thus the kid doing for the credential is just as valid in their choice.'
It is a choice. It is valid. It just also happens to be, well, wasteful ( and maybe a little immoral, but that ship has sailed ) as the kid in question leaves the school with a credential that does not reflect that AND then goes into the world making decisions with a weight of that credential behind him. Thank goodness he is not an actual engineer. Future would look pretty grim then.
<< Just don't claim it's more virtuous than anyone else's hobby, unless you can find a reason.
You may be hanging onto my anecdote, but since that was the only professor at Harvard I had a chance to listen to on the matter, I thought it was relevant and any virtuosity in it is purely coincidental. The point he made was genuinely pretty apt. The point of education at Harvard ( or other notable names ) is not spend your parents money parking your keister for 4 years while waiting for that credential.
<< To call it "wasteful" says that something of "value" is being squandered, but the value is perceived by each of us differently.
No. Just because we perceive things in a certain way, does not automatically mean that there is no objective reality. It just means we don't perceive it ( accurately or otherwise ). This is where I believe this conversation could get interesting, because I think this is what we actually disagree on.
Wasteful is 'expending value carelessly'. Even if we value things differently, using Shelby GT for pizza delivery seems wasteful. I technically have no problem with anyone doing that ( you got money burning your pocket, who am I to judge ), but I am also not going to pretend it is a sensible thing to do.
From where I sit this is not that different from going to Harvard for a credential. Or network. It is just a wasted potential.
And it is sad. For Harvard. Edit: Or society as a whole. I am not sure now.
Ok. I am going to stop here. It is 3am and I clearly typed too much.
> If so, that credential is not only wasteful, it is also worthless.
I think "worth" is a synonym for "value".
> Frankly, at its core, nothing is inherently valuable, because any value is the value we ascribe to things.
I can't reconcile these two statements of yours.
> as we established in previous paragraph, it is a learning institution in name only ( used only for credential )
I've never been, but I imagine there is actually more to it, but some people may not participate in the true learning part. Like the combo meal has fries. I'm just choosing not to eat them.
> The point of education at Harvard ( or other notable names ) is not spend your parents money parking your keister for 4 years while waiting for that credential.
Perhaps Harvard wishes that were not the case, and some of the time it probably isn't. But I find it hard to believe that no one pays tuition for that purpose.
There probably is an objective reality for all know. I even refer to it sometimes. But all I have to work with are these five senses, and this faulty brain.
I ask myself when the time comes that some student will accuse the stuff of libel or slander becuase of false AI plagiarism accusations.
Low light scenarios are just a thing, you would need more expensive hardware do deal with it.
Could it be mostly just be..reality? More expensive hardware doesn't somehow make a darker surface reflect more energy in the visible spectrum. "Low light" is not the same condition as "dark surface in well-lit environment."
Leaving the visible spectrum is one possible solution, but it's substantially more error-prone and costly. This is still not the same solution as classical CV with "more expensive hardware."
Also, this could have been handled much better in the cases that came up in the media if there had been proper human review of all cases before prosecuting the students.
This was at a long-established mall shop that specialized in photography products and services. The same photographer had taken suitable photos of some other people in line ahead of us rather quickly.
The studio area was professional enough, with a backdrop, with dedicated photography lighting, with ample lighting in the shop beyond that, and with an adjustable stool for the subject to sit on.
The camera appeared to be a DSLR with a lens and a lens hood, similar enough to what I've seen professional wedding photographers use. It was initially on a tripod, although the photographer eventually removed it during later attempts.
Despite being in a highly-controlled purpose-built environment, and using photography equipment much better than that of a typical laptop or phone camera, the photographer still couldn't take a suitable photo of this particular woman, despite repeated attempts and adjustments to the camera's settings and to the environment.
Was the photographer "racist"? I would guess not, given the effort he put in, and the frustration he was exhibiting at the lack of success.
Was the camera "racist"? No, obviously not.
Sometimes it can just be difficult to take a suitable photo, even when using higher-end equipment in a rather ideal environment.
It has nothing to do with "racism".
I don't think anyone is saying that the universities or the software companies have some kind of secret agenda to keep black people out. As far as I can tell there's good evidence they're mostly trying to get more black people in (and in some cases to keep Asians out, but that's another story). I also don't think anyone here was acting out of fear or hatred of black people.
What I am claiming is that the universities in question ended up with a proctoring product that was more likely to produce false positives for students with darker skin colors, and did not apply sufficient human review and/or giving people the benefit of the doubt to cancel out those effects. It is quite likely that whatever model-training and testing the software companies did, was mostly on fair-skinned people in well-lit environments, otherwise they would have picked up this problem earlier on. This is not super-woke Ibram X Kendi applied antiracism, this is doing your job properly to make sure your product works for all students, especially as the students don't have any choice to opt out of using the proctoring software beyond quitting their college.
To me it's on the same level as having a SQL injection vulnerability: maybe you didn't intend to get your users' data exposed - about 100% of the time when this happens, the company involved very much did not intend to have a data breach - but it happened anyway, you were incompetent at the job and your users are now dealing with the consequences.
And to the extent that those consequences here fall disproportionately on skin colors (and so, by correlation, on ethnicities) that have historically been disadvantaged, calling this a type of racism seems appropriate. It's very much not the KKK type of racism, but it could very well still meet legal standards for discrimination.
The issue is that, for most people, the term "racism" connotes a moral failing comparable to the secret agendas, fear and hatred, etc. Specifically, an immoral act motivated by a deliberately applied, irrational prejudice.
Using it to refer to this sort of "disparate impact" is at best needlessly vague, and at worst a deliberate conflation known to be useful to (and used by) the "super-woke Ibram X Kendi" types - equivocating (per https://en.wikipedia.org/wiki/Motte-and-bailey_fallacy) in order to attach the spectre of moral outrage to a problem not caused by any kind of malice.
If you're interested in whether someone might have a legal case, you should be discussing that in an appropriate forum - not with lay language among laypeople.
But from the point of view of a black person who has not got a job / college place / tenancy that a comparable white person would have got, I guess it makes sense to say "whatever the cause, I want this problem fixed" and give the symptom rather than the cause the name "racism".
Every major system in the US academic system is aimed to reducing Asian population. It often comes in the guise of DEI with a very wide definition of "Diversity" that rarely includes Asian.
These systems will use subtle features to blackbox racism. They may just be overt and leak over metadata to achieve it, or get smart and using writing styles.
> Despite being in a highly-controlled purpose-built environment
Frankly it sounds like the environment was not purpose-built at all. It was built to meet insufficient standards, perhaps.
LLMs already discriminates against African-American English. You could argue a human grader would as well, but all tested models were more consistent in assigning negative adjectives to hypothetical speakers of that dialect.
NB: I am not African-American, nor did I grew up on an African-American community, and I performed very well on all sorts of verbal tests. Yet, even I made the all intensive purposes mistake until well into adulthood. Probably a Midwestern thing.
(Makes me worry I'm not paying enough attention, that I can't).
So far. Unless there is a new generation of teachers who are no longer able to learn on non-AI generated texts because all they get is grammatically corrected by AI for example...
Even I am using Grammarly here (as being non-native), but I usually tend to ignore it, because it removes all my "spoken" style, or at least what I think is a "spoken style"
So do humans. Many people have pet phrases or words that they use unusually often compared to others.
But as a second language learner, you notice that people get stuck on particular words during writing sessions. If I run into a very unusual (and unnecessary) word, I know they're going to use it again within a page or two, maybe once after that, then never again.
I blame it on the writer remembering a cool word, or finding a cool word in a thesaurus, then that word dropping out of their active vocabulary after they tried it out a couple times. There's probably an analogue in LLMs, if just because that makes unusual words more likely to repeat themselves in a particular passage.
I think words I've used recently are easier to access, as if there's a cache for items recently retrieved from deeper layers of memory.
It would also identify random 5-6 word phrases and attribute them to different random texts on completely different topics where those same 5 words happened to appear.
It seems likely that a suitably disciplined student could look a lot like ChatGPT and the cost of a false accusation is extremely high.
Dodging the detector is done regardless of whether or not one has used AI to write that paper.
That’s the problem. It is trivially easy, 99% of the time. But that misses the entire point of the article.
If I got 99% on an exam I’d say that was trivially easy. But making one mistake in a hundred is not ok when it’s someone else’s livelihood.
"You asked us to write an essay on the Civil War. The first thing I did was ask an AI to explain it to me, and I asked the AI some follow-up questions. Then I did some research using other sources and wrote my paper."
It might even be a true story, and in such a case it's not surprising that the student would repeat words they encountered while studying.
For a human who deals with student work or reads job applications spotting AI generated work quickly becomes trivially easy.
I couldn't cheat in high school because we couldn't use our phones during class. Not for worksheets nor quizzes and especially not exams whether they be multiple choice, oral, or essays.
Yet the top threads here act like we need a whole refactor of schooling, many people suggesting we rely on viva voce exams and proctored exams. What exactly do you think that's solving over a simple classroom scantron test where the teacher ensures people aren't on their phones?
In many places, particularly in the U.S., there are few invigilated exams, and quite a lot of your overall grade will be comprised of coursework. This, combined with the inexorable advance of digitalisation of education has led to where we are now.
Certainly once you get to university level, there are projects which simply take too long to be done in the classroom, such as a dissertation or final report. These projects have always been vulnerable to commissioning rather than plagiarism, and you’d be appalled to realise how common it actually is even in higher prestige places. LLMs have simply lowered that bar to make it even more common.
This is a genuine problem, and people are more sophisticated cheaters than you might initially think.
This was before generative AI became so commonplace, and I got the impression this is super common place. It was a really disillusioning moment for me.
Our institutions are failing us, and I have never been more disillusioned.
That's what proctoring is.
In either case, we need to change our standards around mastery of subject matter.
Judges and police officers arent 100% accurate too
For a supposedly educated and thinking person like a professor, if they don't understand "AI" and can't reason that it can most certainly be wrong, they just shouldn't be allowed to use it.
Threatening someone like the people in the article with consequences if they're flagged again, after false flags already, is barbaric; clearly the tool is discriminating against their writing style, and other false flags are probably likely for that person.
I can't imagine what a programming-heavy course would be like these days; I was once accused alongside colleagues of mine (people I'd never spoken to in my life) of plagiarism, at university, because our code assignments were being scanned by something (before AI), and they found some double-digit percentage similarity, but there's only so many ways to achieve the simple tasks they were setting; I'm not surprised a handful out of a hundred code-projects solving the same problem looked similar.
My comment from a few days ago.
The origin was a conversation with a girl who said she'd been pulled into a professor's office and told she was going to be reported to whatever her university's equivalent of Student Conduct and Academic Integrity is over using AI - a matter of academic honesty.
The professor made it clear in the syllabus that "no AI" was allowed to be used, spent the first few days of class repeating it, and yet, this student had been assessed by software to have used it to write a paper.
She had used Grammarly, not ChatGPT, she contended. They were her words and ideas, reshaped, not the sole product of a large language model.
In a world where style suggestion services are built into everything from email to keyboards, what constitutes our own words? Why have ghostwritten novels topped the NYT Best Sellers for decades while we rejected the fitness of a young presidential hopeful over a plagiarized speech?
Integrity doesn't exist without honesty. Ghostwriting is when one person shapes another person's truth into something coherent and gives them credit. A plagiarized speech is when someone takes another person's truth as their own, falsely. What lines define that in tools to combat the latter from the former, and how do we communicate and enforce what is and isn't appropriate?
Writing essays isn't just about your ideas. It's also a tool to teach communication skills. The goal of an essay isn't to produce a readable paper, until you start your PhD at least; it's to teach a variety of skills.
I don't really care about the AI generated spam that fills the corporate world because corporate reports are write-only anyway, but you can't apply what may be tolerated in the professional world to the world of education.
It's still not cheating. English assignments aren't about the practice of writing English, you stop doing that in primary school. It's analysis of English texts in which people have been using spelling and grammar checkers since their inception. It's not even cheating to have someone proofread and edit your paper, it's usually encouraged, and Grammarly is just a worse-than-human editor.
Saying "AI" becomes meaningless if we're all using it to mean different things. If I use computer vision to perform cell counts, or if an ESL student uses deepl to help translate a difficult to express idea, would we be in breach of student conduct?
The real answer is "ask your professor first", but with how second nature many of these tools have become in P12 education, it may not occur to students that it might be necessary to ask.
Whoops, with that little comment I suspect you've invalidated most English papers written in the past 2 decades. Certainly all of mine! Thanks spellcheck.
But I think it's actually not all that different, particularly in the context of "essays teach writing." It used to be human work to analyze sentences for passive voice, remember the difference between there/their/they're, and understand how commas work, but now the computer handles it.
(Relevant sidenote: Am I using commas correctly here? IDK! I've never fully internalized the rules!)
People keep using these "gotcha" examples and never actually look at the stats for it. I get it, there are some terrible detectors out there, and of course they are the free ones :)
https://edintegrity.biomedcentral.com/articles/10.1007/s4097...
GPTZero was correct in most scenarios where they used basic prompts, and only had one false positive.
We did a comparison of hand reviewed 3,000 9-12th grade assignments and found that GPTZero holds up really well.
In the same way that plagiarism detectors need a process for review, your educational institution needs the same for AI detection. Students shouldn't be immediately punished, but instead it should be reviewed, and then an appropriate decision made by a person.
> GPTZero was correct in most scenarios where they used basic prompts, and only had one false positive.
One false positive out of only "five human-written samples", unless I'm misreading.
Say 50 papers are checked, with 5 being generated by AI. By the rates of GPTZero in the paper, 3 AI-generated papers would be correctly flagged and 9 human-written papers would incorrectly flagged. Meaning a flagged paper is only 25% likely to actually be AI-generated.
Realistically the sample size in the paper is just far too small to make any real conclusion one way or another, but I think people fail to appreciate the difference between false positive rate and false discovery rate.
Institutions have to enforce rules around these things, if they do not within 10 years their degrees will be pointless.
It's what happens when you believe someone to have cheated that matters. If it's not blatant cheating, then you cannot punish them for it. These tools exist to catch only the worst offenders.
These limited time-window assessments are also (a) artificial (don't always reflect how the person might use their knowledge later) (b) stressful (some people work better/worse with a clock ticking) and (c) subject to more variability due to the time pressure (what if you're a bit sick, or have had a bad day or are just tired during the time window?).
AI makes it impossible to rely on out-of-class assignments to evaluate the kids' knowledge. How we respond to that is unclear, but relying on cheating detectors is not going to work.
Teacher explains material, you get homework about the material and are graded on it.
It shouldn't be like that. If the work (i.e. the exercises) are important to grasp the material, they should be done in class.
Also removes the need of hiring tutors.
I'd like to offer what I've come to realize about the concept of homework. There are two main benefits to it: [1] it could help drill in what you learned during the lecture and [2] it could be the "boring" prep work that would allow teachers to deliver maximum value in the classroom experience.
Learning simply can't be confined in the classroom. GP suggestion would be, in my view, detrimental for students.
[1] can be done in class but I don't think it should be. A lot of students already lack the motivation to learn the material by themselves and hence need the space to make mistakes and wrap their heads around the concept. A good instructor can explain any topic (calculus, loops and recursion, human anatomy) well and make the demonstration look effortless. It doesn't mean, however, that the students have fully mastered the concept after watching someone do it really well. You only start to learn it once you've fluffed through all the pitfalls at least mostly on your own.
[2] can't be done in class, obviously. You want your piano teacher to teach you rhythm and musical phrasing, hence you better come to class already having mastered notation and the keyboard and with the requisite digital dexterity to perform. You want your coach to focus on the technical aspects of your game, focus on drilling you tactics; you don't want him having to pace you through conditioning exercises---that would be a waste of his expertise. We can better discuss Hamlet if we've all read the material and have a basic idea of the plot and the characters' motivations.
That said, it might make sense to simply not grade homeworks. After all, it's the space for students to fail. Unfortunately, if it weren't graded, a lot of students will just skip it.
Ultimately, it's a question of behavior, motivation, and incentives. I agree that the current system, even pre-AI, could only barely live up to ideals [1] and [2] but I don't have any better system in mind either, unfortunately.
I fundamentally disagree - I vividly remember, many times during homework in maths for example, I realised that I am stuck and so don’t understand something explained earlier, and I need to ask someone. For me, my parents were able to help. But later in Highschool, when you get to differential equations - they no longer can. And obviously if your parents are poorly educated they can’t rather.
Second point, there is no feedback loop this way - a teacher should see how difficult is his homework, how much time students spend on it, and why they are struggling. Marking a piece of paper does not do it. There was wild inconsistency between teachers for how much homework they would set and how long they thought it would take students.
Lastly, the school + homework should be able to accommodate tag the required learning within 1 working day. It is anyway a form of childcare while parents work
It’s not when you reframe it in Puritanical terms. Keep the children busy for 12 hours per day: If they get some practice on their courses, great, but busy, quiet children won’t fall in with the devil.
I wish I could get a refund on all the wasted childhood I spent doing useless homework on subjects I have not used since. No, it didn’t make me “a well-rounded person,” it just detracted from the time I could spend learning about computers—a subject my school could not teach me.
- placing less emphasis on numerical grades to disincentive cheating (hard to measure success) - open response written questions (harder to teach, harder to grade) - reading books (hard to determine if students actually did it) - proof based math (hard to teach)
Instead we keep imagining more absurd surveillance systems “what if we can track student eyes to make sure they actually read the paragraph”
but somehow, we don't trust teacher anymore. Those in power want to check that the teacher actually makes his job so they want to see wome written, reviewable proof... So the grades are there both to control the student and the teacher. WWW (What a wonderful world).
On the other hand, kids do blindly use the hell out of ChatGPT. It's a hard call: teach to the cheaters or teach to the good kids?
I've landed on making take-home assignments worth little and making exams worth most of their grade. I'm considering making homework worth nothing and having their grade be only 2 in-class exams. Hopefully that removes the incentive to cheat. If you don't do homework, then you don't get practice, and you fail the two exams.
(Even with homework worth little, I still get copy-pasted ChatGPT answers on homework by some students... the ones that did poorly on the exams...)
I'd be cautious about that, because it means the kids with undiagnosed ADHD who are functionally incapable of studying without enforced assignments will just completely crash and burn without absorbing any of the material at all.
Or, at least, that's what happened to me in the one and only pre-college class I ever had where "all work is self-study and only the tests count" was the rule.
My second and third semesters went exactly as you described for courses where I was exposed to new things and wasn't just repeating high school - mainly because I had no training or coping mechanisms for learning under that type of pedagogy.
After getting my ass kicked in exams and failing a class for the first time in my life, I finally grokked that optional homework assignments were the professor's way of communicating learning milestones to us, and that even though the professor said they weren't graded (unless you asked), you still had to do them or you wouldn't learn the material well enough to pass the exam.
Still had a few bad grades because of the shit foundation I built for myself, but I brought a 2.2 GPA up to a 3.3 by the end.
The point is that it takes is exposure to that style of teaching before it can really be effective.
I feel like this is almost exactly moving all evaluation into the class. If "little" becomes nothing, it is exactly that.
I feel this was always the best strategy. In college, how much homework assignments were worth was an easy way to evaluate how bad the teacher was and how lightweight the class was going to be. My best professors dared you not to do your homework, and would congratulate you if you could pass their exams without having done it.
The very best ones didn't even want you to turn it in, they'd only assign problems that had answers in the back of the book. Why put you through a entire compile cycle of turning it in, having a TA go over it, and getting it back when you were supposed to be onto the next thing? Better and cheaper to find out you're wrong quickly.
I'm considering making homework worth nothing and having their grade be only 2 in-class exams.
- Proctored exams at the end of 2 years of study (the last 2 years of high school)
- Proctored exams at the end of 2 years of study (the last 2 years of university)
There are more students than ever, and lots of schools now offer remote programs, or just remote options in general for students, to accommodate for the increased demand.
There's little political will to revert to the old ways, as it would drive up the costs. You need more space and you need more workers.
The teacher accuses the student of using the LLM to perform the task they are assigned. This entails not properly understanding the assignment and presenting an accomplishment which has not been achieved by the student themselves.
On the other hand the teacher using an LLM tool also do not understand the reasoning of the decision and present often present them as their own judgement. A judgement that has not truly been felled by the teacher because they do not use the tool for understanding but for deferring their responsibilities.
In doing so the teacher is engaging in the same act of (self-)deception they are accusing the student of: presenting an achievement not truly reached through their own understanding, even if the situation necessitates it (non-deferrable learning vs. non-deferrable decision).
The use of LLM-detection in this way thus mirrors the very problem it seeks to address.
Don't know why these companies are spending so much developing this technology, when their customers clearly aren't checking how well it works.
Plagiarism detectors kinda work, but you can always use one to locate plagiarized sections and fix them yourself.
I have a plagiarism rate under 5%, usually coming from the use of well known phrases.
An AI usually has over 10%.
Obviously that doesn't help in an academic context when people mark their citations.
The perplexity checks don't work, as humans seem to vary highly in that regard. Some of my own text has less perplexity as a comparable AI text.
I did have a very interesting case once of a student who copied and pasted someone's Master's thesis for sections of her paper, but also listed that thesis in the citations... it remains up to the jury (not me) to decide whether she just didn't understand what plagiarism was. I would not have known if Turnitin didn't mark it as 30% plagiarized.
Disclaimer: Someone more senior than I was in charge of the decision to use this software, but it was interesting to see it in action
So sure; not free, but free as in beer.
Uh huh. Except for the very very very high frequency with which they weren't.
Also, people traded favors. I'll grant you that's not technically "free," but it may as well have been.
I work with people in their 30s That cannot write their way out of a hat. Who cares if the work is AI assisted or not. Most AI writing is super dry, formulaic and bad. The student doesn’t recognize this the give them a poor mark for having terrible style.
Actually being able to generate good text is entirely separate evaluation. And AI might have place there.
LLM can generate text that is as entertaining and whimsical as its training dataset gets with no effort on your side
I say good riddance, that's exactly how it should be. At-home busywork is a scourge on especially K-12 students. Yet, every teacher has been loading their students up with homework, because that's their idea of what a "good teacher" is supposed to do.
The faster technology overcomes this problem, the better.
Helping kids find value in education is the only important concern here and adding an AI checker doesn't help with that.
Exactly. It also does the complete opposite. It teaches kids from fairly early on that their falsely flagged texts might as well be just written with AI, further discouraging them from improving their writing skills. Which are still just as useful with AI or not.
Models before 2017-2018 (first gpt/bert) didn’t produce any decent text, and before gpt2/gpt3 (2020) you wouldn’t get an essay-grade text.
So you need to go back only 4-5 years to be certain an essay didn’t use AI.
Obviously we will go back to in class writing.
Oh... It is the story of my school math education. I always got bad marks, because I was "too stupid to come up with this particular solution to the problem". I didn't thought it was really unfair, because I thought myself to be lazy, and I looked for such solutions to math problems that would minimize my work. Oftentimes I ignored textbook ways to solve problems and used my own. I believed that it was a cheating, so naturally I got worse marks, but I put up with that, because I was lazy to do it in more complex way from a textbook.
That would be a pretty sad outcome. In my high school we did both in-class essays and homework essays. The former were always more poorly developed and more more poorly written. IMO students still deserve practice doing something that takes more than 45 minutes.
What may seem obvious based on earlier-era measures of student comprehension and success is not the case in many schools anymore.
Look up evidence based grading, equitable grading, test retake policies, etc.
I think the answer is "The Stakes" of getting a poor "grade" that follows you in and ranks you. Eliminate that, and tests become a valuable self assessment of where a student is. Teaches become partners in growth, not adversaries that can cause long term harm with a black mark.
Universities and teachers will need to adjust to the reality that this stuff is here to stay. There's some value in learning how to write properly, of course. But there are other ways of doing that. And some of those ways actually involve using LLMs to criticize and correct people's work instead of having poor teachers do that.
I did some teaching while I was doing a post doc twenty years ago. Reviewing poorly written student reports isn't exactly fun and I did a fair bit of that. But it strikes me how I could use LLMs to do the reviewing for me these days. And how I could force my students to up their standards of writing.
These were computer science students. Most of them were barely able to write a coherent sentence. The bar for acceptable was depressingly low. Failing 90% of the class was not a popular option with either students or staff. And it's actually hard work reviewing poorly written garbage. And having supported a few students with their master thesis work, many of them don't really progress much during their studies.
If I were to teach that class now, I would encourage students to use all the tools available to them. Especially AI. I'd set the bar pretty high.
The point of essays is not to have essays written. The teacher already knows. The point is to practice putting together a coherent thought. The process, not the product, is a the goal.
Eventually we'll come up with a way to demonstrate that along with, rather than despite, AI. But for the moment we have machines that can do the assignment much better than students can, and the students won't get any better if they let the machine do all of the work.
For additional context the short essay format as an evaluation tool is very much a Anglo-saxon university form factor.
There are several other cultures in the world, in particular stemming from Latin/Francophone school of thought, in the old 'cathedra' style university where students are either subjected to written exams only or even historically (less so nowadays) also 'oral' exams (Oratory not dental exams).
There are studies showing up already that students educated with LLMs end up retaining nothing.
How would you set the bar pretty high? How would you avoid just evaluating ChatGPT, instead of the actual student?
And of course if all papers are up to standards on that (which IMHO would be a massive improvement already from an educational point of view), you'd be looking for other criteria to judge the papers that maybe showcase things that are actually of value. Like problem solving, critical thinking, originality, etc. I'd be looking for signs of the student having a good grip on the subject matter.
Perhaps I'd do a little verbal exam. I might grill them a bit on the subject they wrote about and make sure they understand what they submitted. Somebody that did the leg work of coming up with something good and that did the research, would be able to answer questions about it and be able to discuss the key points. Ask them some questions about other work they are referencing. Etc.
I just think trying to keep students from using tools that are out there is a lost cause.
If I was still in school I would write my docs in a Google Doc which provides the edit history. I could potentially also record video of me typing the entire document as well or screen recording my screen.
“After her work was flagged, Olmsted says she became obsessive about avoiding another accusation. She screen-recorded herself on her laptop doing writing assignments. She worked in Google Docs to track her changes and create a digital paper trail. She even tried to tweak her vocabulary and syntax. “I am very nervous that I would get this far and run into another AI accusation,” says Olmsted, who is on target to graduate in the spring. “I have so much to lose.”
That said, I'd imagine if someone resorts to using generative text their edits would contain anomalies that someone legitimately writing wouldn't have in terms of building out the structure/drafts. Perhaps that in itself could be auto detected more reliably.
In the demoscene, still graphics competitions, at least as far as I remember, most organizers defend against cheating by requiring the artists to capture snapshots of their work over time to show that they didn't:
1) steal other people's graphics
2) just use some kind of tool to convert a photo or rework somebody else's work
During the presentation of the works for voting, all of the stages are typically displayed for the crowd.
This works today because AI tools typically don't output any intermediate steps, and if they do they don't look anything like what a human would produce.
This works for most educational assignments as well.
Heck, in fields where Microsoft Word is an option, Sharepoint preserves the change history and it's a pretty simply matter to just review that history to show progression, edits over time, and all the other elements you might think of to show that the student actually wrote the document themselves. It also helps frustrate people who might just copy-paste dump other work into the document. The teacher doesn't need to review or grade ever single revision, just have it all accessible.
Two other practical examples of where this has worked:
1) In university all of my mathematics professors required us to "show our work", which helped with partial credit in cases we arrived at the wrong answer, but also defeated the use of advanced symbolic systems that simply barfed out the answer.
2) At work I had an issue with an employee who I suspected was claiming other people's work. He had a role that was supposed to be reviewing and editing their documents so it was difficult to prove. However, a review of Sharepoint's edit history for multiple documents showed no edits made by him on several major documents. This sparked an inquiry to ensure he wasn't using some alternative method, and the rest of was simple to deal with HR.
It seems like a long term loosing proposition.
Sounds like a good candidate to IPO early
"The Elite College Students Who Can’t Read Books"
https://www.theatlantic.com/magazine/archive/2024/11/the-eli...
"I read Hamilton this month," means heard the audiobook while commuting in the car or on the train.
So now they can all read books again.
For those going to college, I strongly advise picking a department where such scanning is not performed.
For those in public school, sue.
My take is, if she used AI to generate that, she didn't use a very good one. I don't think ChatGPT would make the grammar and clarity mistakes that you see in the image text.
I see this:
"should be exposed to many of these forms and models to strengthen understanding" - much better as "should be exposed to as many of these forms and models as possible to strengthen their understanding"
"it is mentioned that students should have experiencing understanding the..." - plainly wrong, better would be "it is mentioned that students should have experience understanding the..."
"time with initial gird models" -> "time with initial grid models"
And there are other lines that could be improved.
My opinion is, the only solution to this problem is to allow AI detectors to flag work, but that when a work is flagged, that flagging just triggers a face to face meeting between the student and the professor, where the student is required to show through discussion of the work that they understand it well enough to have written it.
However! Often the professor is too busy, or isn't smart enough to review the writing of the student carefully enough to determine whether the student really wrote it. What to do? Why of course: invent AI systems that are really good at interviewing students well enough to tell if they really wrote a piece of work. Yeah you laugh but it will happen some day soon enough.
I don't wanna trot out "think of the children" bullshit here but it's hard for me to not notice that this trend has been happening since smartphones became normal and schools have increasingly become utterly toothless with regard to enforcing standards in education, i.e. "you need to know this shit in order to move to the next grade up." Nobody does that anymore. Just fudge the scores with extra credit or make-up assignments and send them up the chain to be a different teacher's problem next year.
> My opinion is, the only solution to this problem is to allow AI detectors to flag work, but that when a work is flagged, that flagging just triggers a face to face meeting between the student and the professor, where the student is required to show through discussion of the work that they understand it well enough to have written it.
You said it yourself in the subsequent paragraph, but if professors had this much time and energy to teach, their kids wouldn't be writing like deprecated GPT instances in the first place. We need to empower teachers and schools to fail children so they can be taught and experience consequences for lack of performance, and learn to do better. They have no reason to try because no one will hold them accountable, personally or systemically. We just let them fail and keep failing until they turn into failures of adults living in their parents basements playing Elden Ring all day and getting mad at each other over trivial bullshit on social media.
I work in a field where I _think_ that clarity of communication is critically important. But then I see people that can't read or write worth a damn getting promotions and the like, and I think, maybe I'm just an old curmudgeon.
To others (not you ToucanLoucan): If you are reading this and you are wondering how you can become the person who doesn't send the email that is taken to mean "turn the server off now" when what you meant to say was "turn the server on now", all you really need to learn is to fully proof read your messages before you click send, in my opinion. And to write as much as you can. Everything else will take care of itself; you will naturally get better and better at it.
I agree in general that we need to have higher standards but that complaint predates smartphones by decades. One of the big challenges here is that consistency was all over the place historically but we have better measurements now and higher expectations for students, and some of the cases where students were allowed to slide were misguided but well-intended attempts to mitigate other problems – for example, many standardized tests had issues with testing things like social norms or English fluency more then subject matter literacy so there’s a temptation to make them less binding when it should be paired with things like improving tests or providing ESP classes so, say, a recent immigrant’s math score isn’t held down by their ability to read story problems.
One other thing I’d keep in mind is that this is heavily politicized and there are massive business conflicts of interest, so it’s important to remember that the situation is not as dire as some people would have you believe. For example, PISA math scores are used to claim Americans are way behind but that’s heavily skewed by socioeconomic status and tracking in some other countries – when you start adjusting for that, the story becomes less that American students as a whole are behind but rather that our affluent kids are okay but we need to better support poor kids.
This is an ethics problem (people willing to cheat), this is a multi cultural problem (different expectations of what constitutes cheating) this is an incentive problem (credentialism makes cheating worth it).
Those are hard problems. So a little tech that might scare students and give the professor a feeling of control is a band aid.
I wager we will soon see the same for text. Automatic insertion of the right amount of believable mistakes will become a thing.
The pandora's box has been opened with regards to large language models.
For me it was the latter. Luckily, where I grew up in India the bogeyman was entrance exams. They are bad but in a way better than essays because you have a very clear expectation of success.
Either way, I hope GenAI finally makes essay writing obsolete so that we may move on to other better methods of assessing students. For those in flux as this situation rolls through - my sympathies. Educators have been lazy and you're paying the price.
I have absolutely no problem using it in a more advanced field where the basics are already done and the focus is on research, for example, but at lower levels I'd likely consider it dishonest.
She loves it. It doesn’t cause her to be any less attentive to her writing; it just makes it possible to write.
I was not really referring to accommodations under the ADA. For people that do not require accommodations, the use of them is unfair to their classmates and can be detrimental to their ability to perform without them in the future, as there is no requirement to have the accommodations available to them. This is not the case for someone with dyslexia.
So if anyone knows a good tool, that is flexible enough to support proper writing and able to run locally on a machine, hints appreciated.
If I could short higher education, I would. Literally all its foundational principles are bordering on obviously useless in the modern world, and they keep doubling down on the same fundamentals (a strict set of classes and curriculum, almost complete separation of education with working experience, etc), only adapting their implementation somewhat.
Either way, this is a giant lawsuit waiting to happen. Schools need to ban these tools asap. They will never work and anyone who takes them seriously… I have a dousing rod that can detect AI available for 29.95
If you want more proof, then you can take the essay, give it to chatGPT and say, "Please give me a report showing how this essay is written to en by AI."
People treat AI like it's an omniscient god.
And ChatGPT will happily argue whichever side you want to take. I just passed it a review I wrote a few years ago (with no AI/LLM or similar assistance), with the prompts "Prove that this was written by an AI/LLM: <review>" and "Prove that this was written by a human, not an AI/LLM: <review>", and got the following two conclusions:
> Without metadata or direct evidence, it is impossible to definitively prove this was written by an AI. However, based on the characteristics listed, there are signs that it might have been generated or significantly assisted by an AI.[1]
> While AI models like myself are capable of generating complex and well-written content, this specific review shows several hallmarks of human authorship, including nuanced critique, emotional depth, personalized anecdotes, and culturally specific references. Without external metadata or more concrete proof, it’s not possible to definitively claim this was written by a human, but the characteristics strongly suggest that it was.[2]
How you prompt it matters.
[1] https://chatgpt.com/share/67164ec9-9cbc-8011-b14a-f1f16dd8df...
[2] https://chatgpt.com/share/67164ee2-a838-8011-b6f0-0ba91c9f52...
For an assignment completed at home, on a student's device using software of a student's choosing, there can essentially be no proof. If the situation you describe becomes common, it might make sense for a school to invest into a web-based text editor that capture keystrokes and user state and requiring students use that for at-home text-based assignments.
That or eliminating take-home writing assignments--we had plenty of in-class writing when I went to school.
According to an undergraduate student who babysits for our child, some students are literally screen recording the entire writing process, or even recording themselves writing at their computers as a defense against claims of using AI. I don't know how effective that defense is in practice.
That, or the uni shall give me a separate machine to write on, only for that purpose.
Well yes, in-person proctored is the gold standard. For those who can’t or won’t go in person, something invasive is really the only alternative to entirely exam-based scoring.
So no, you don’t exactly get a trial by a jury of your peers, but it isn’t like they are averse to evidence being presented.
This evidence would be fairly trivial to refute, but I agree it is a burden no student needs or wants.
They issue the claim, the judgement and the penalty. And there is nothing you can do about it.
Why? Because they *are* the law.
You can sue the university, and likely even win.
They literally are not the law, and that is why you can take them to court.
A kid living in a wealthy Boston suburb used AI for his essay (that much is not in doubt) and the family is now suing the district because the school objected and his chances of getting into a good finishing school have dropped.
On the other hand you have students attending abusive online universities who are flagged by their plagiarism detector and they wouldn't ever think of availing themselves of the law. US law is for the rich, the purpose of a system is what it does.
You don’t need to be rich to change the law. You do need to be determined, and most people don’t have or want to spend the time.
Literally none of that changes the fact that the Universities are not, themselves, the law.
But that they can be sued in a court of law is actually a very big deal; it is the defining thing that makes them not the law.
A reminder of what I was responding to: “They issue the claim, the judgement and the penalty. And there is nothing you can do about it. Why? Because they are the law.”
That is plainly untrue. There is something you can do about it. You can sue them, precisely because they are not the law.
They’re not. That doesn’t make it less stressful, annoying, or unnecessary to fight them.
Police aren't the law because they have been sued?
Your police argument is a strawman.
Then, if some assignment has a different writing style, not only could that potentially detect more simple uses of AI, but it might detect the old trick of getting their friend to give them a copy of last year's assignment and passing off that work as theirs, since their friends writing style would be different.
Of course, if the student is smart enough to train the AI on their own writing style, this might not work so well.
But it might help get a guide for people who naturally write in a way that will get flagged by these tools, such as Neurodivergent people and hopefully prevent them from being falsely accused, since it would already be known from the start that this is their natural writing style.
This is waay more common and just impossible to catch. The only students caught here are those that put no effort in at all
If one can "rewrite it from scratch" in a way that's actually coherent and gets facts correct, then they learned the material and can write an original paper.
> This is waay more common and just impossible to catch.
It seems a good thing that this is more common and, naturally, it would -- perhaps should, given the topic -- be impossible to catch someone cheating when they're not cheating.
(“Thorough examination” as a term is too long though — let’s just call them “thors”.)
—
In seriousness the above only really applies at University level, where you have adults who are there with the intention to learn and then receive a final certification that they did indeed learn. Who cares if some of them cheat on their homework? They’ll fail their finals and more fool them.
With children though, there’s a much bigger responsibility on teachers to raise them as moral beings who will achieve their full potential. I can see why high schools get very anxious about raising kids to be something other than prompt engineers.
There's nothing moral about busywork for busywork's sake. If their entire adult life they'll have access to AI, then school will prepare them much better for life if it lets them use AI and teaches them how to use it best and how to do the things AI can't do.
I even knew those who did the work honestly, received high marks, and then couldn't actually write reasonable code. In my capstone project one of my teammates ask me if his code needed to compile or not. Another couldn't implement a function that translates ASCII letters -> numbers without a lookup table.
Anyway, all of this to say, maybe we just shouldn't care about grades as much.
Writing their essays by hand. That is what they are doing.
But overall as someone who teaches a computational class where students write code, I'm still at loss on how to deal with ChatGPT-cheating.
So far I'm using 1) referring to lecture notes when discussing possible methods to solve problems 2) converting problems to images to make it harder to copy-paste into chatgpt 3) Being stricter for code that is almost correct but does not run 4) Using hidden datasets, or input data-files that needs to be processed to figure the best solution
But it's all a bit futile I feel.
Amusingly, you can push ChatGPT type systems into other styles of writing.
I put in the preamble to the US constitution and asked for different styles:
Modern:
We, the people of the United States, come together to build a stronger, more united country. We want to create fairness for everyone, keep peace at home, ensure our safety, support each other’s well-being, and protect our freedoms for ourselves and future generations. This is why we are establishing this Constitution for the United States.
Gun nut:
We the People of the United States, rallying together to create a stronger, more unified nation, are all about upholding justice, keeping the peace at home, and defending our rights. We’re here to support one another, ensure our freedoms, and safeguard the blessings of liberty for ourselves and future generations. That’s why we’re putting this Constitution in place for the United States of America—because we believe in freedom, security, and our right to bear arms!
Woke:
We, the people of the United States, are coming together to create a more inclusive and equitable society. Our goal is to establish justice for all, promote peace within our communities, ensure safety for everyone, and uplift the well-being of every individual. We are committed to protecting our freedoms and rights, not just for ourselves, but for future generations. This is why we are enacting this Constitution for the United States of America, to foster a nation that values diversity, equality, and the dignity of all.
Christian:
We, the people of the United States, gather together in faith and unity to build a more perfect nation under God. Our purpose is to establish justice and promote peace in our homes and communities. We strive to protect our common security and care for one another’s well-being, guided by the principles of love and compassion. In doing so, we seek to uphold the blessings of liberty that God has granted us, not just for ourselves, but for future generations. This is why we commit to this Constitution for the United States of America, as we seek to honor His will in all that we do.
If you're having trouble with essays being detected as AI, you can run them through an LLM to rewrite them in a different tone.
We, the people of the United States, unite to forge a more harmonious society, guided by the principles of collective strength and shared responsibility. Our aim is to establish justice and maintain social stability, ensuring the well-being of all citizens through solidarity and mutual support. By safeguarding our national defense and promoting the common welfare, we embody the spirit of unity and progress. This commitment to our foundational values is why we establish this Constitution for the United States of America, as we work toward a prosperous future for all.
MAGA:
We, the great people of the United States, are coming together to create a stronger, more united nation. Our mission is to establish true justice, ensure peace in our communities, and defend our country with pride. We’re all about promoting the well-being of hard-working Americans and securing our freedoms for ourselves and future generations. That’s why we’re affirming this Constitution for the United States of America—because we believe in making our nation great and preserving our rights for all!
Martin Luther King:
We, the people of the United States, come together in a spirit of unity and purpose, striving to create a more perfect union. It is our sacred duty to establish justice and ensure tranquility in our communities, uplifting one another in love and understanding. We must provide for the common defense and promote the welfare of all, recognizing that true freedom is a collective journey. In this endeavor, we seek to secure the blessings of liberty for ourselves and future generations. Therefore, we affirm this Constitution for the United States of America as a testament to our commitment to equality, justice, and the enduring hope for a brighter tomorrow.
Trump:
We, the amazing people of the United States, are coming together to build a tremendous nation, folks. Our goal? To create a more perfect union—believe me, it’s going to be fantastic! We’re establishing real justice, keeping our communities safe, and defending our great country. We’re promoting the welfare of all hardworking Americans and securing our incredible freedoms for ourselves and our future generations. That’s why we’re putting this Constitution in place for the United States of America—because we’re making America great again, and nobody does it better!
ChatGPT has automatic blithering nailed.
Classifiers sometimes produce false positives and false negatives. This is not news to anyone who has taken a ML module. We already required students back then to be able to interpret the results they were getting to some extent, as part of the class assignment.
Even before AI detectors, when Turnitin "classic" was the main tool along with JPlag and the like, if you were doing your job properly you would double-check any claims the tool produced before writing someone up for misconduct. AI detectors are no different.
That said, you already catch more students than you would think jut by going for the fruit hanging so low it's practically touching the ground already:
- Writing or code that's identical for a large section (half a page at least) with material that already exists on the internet. This includes the classic copy-paste from wikipedia, sometimes with the square brackets for references still included.
- You still have to check that the student hasn't just made their _own_ git repo public by accident, but that's a rare edge case. But it shows that you always need a human brain in the loop before pushing results from automated tools to the misconduct panel.
- Hundreds of lines of code that are structurally identical (up to tabs/spaces, variable naming, sometimes comments) with code that can already be found on the internet ("I have seen this code before" from the grader flags this up as least as often as the tools).
- Writing that includes "I am an AI and cannot make this judgement" or similar.
- Lots of hallucinated references.
The future in this corner of the world seems to be a mix of
- invigilated exams with no electronic devices present
- complementing full-term coding assignments with the occasional invigilated test in the school's coding lab
- students required to do their work in a repo owned by the school's github org, and assessing the commit history (is everything in one big commit the night before the deadline?). This lets you grade for good working practices/time management, sensible use of branching etc. in team projects, as well as catching the more obvious cases of contract cheating.
- viva voce exams on the larger assignments, which apart from catching people who have no idea of their own code or the language it was written in, allows you to grade their understanding ("Why did you use a linked list here?" type of questions) especially for the top students.This strikes a chord !
When it flags it, it’s 100% correct. There are no false positives. It’s still possible to have false negatives.
Btw if anyone is recruiting and interested in such a tool, contact me. Details in profile.
>> I modified my written screen in a way that ...
What's a written screen? (I could not find via a quick Google search.)
Students already have the entire internet at their disposal, the wealth of all human knowledge right in their hands, it's been over an entire internet generation at this point. I'd go as far as to argue many courses on YouTube are much higher quality than what they recieve in a classroom. Are students learning more than ever?
No, they are not, in fact many argue it's worse than ever, English comprehension and writing have regressed significantly.
Students often need support from their teachers, teachers often are more present than their parents. It simply isn't the case that most people will be a self-taught learner with an AI.
I do agree educators should pivot to more hands on, non-writing lessons such as debates, instead of written papers, but we're not going to improve writing skills without having written papers..
My middle school English classes were dominated by sentence diagramming - I'm not sure that's taught anymore. I hated it at the time but damn if it wasn't effective.
I know that wouldn't fly but it would be interesting to see something like that.
Learning centers need to adopt similar principles in order to avert overt homework and exam cheating.
Do we trust the students, or the professors? No.
So why continue to treat them as if we did?
It's because AI detectors don't have 100% accuracy that they are considered bad.
Working AI detectors are bad.
False cheating accusations are collateral damage to those naive enough to participate.
Post-apocalyptic education
What comes after the Homework Apocalypse
by Ethan Mollick
Clearly, something other than education is going on.
AI isn't going to help if you ask it the wrong questions.
In practice, reconsidering how clsssed are structured is a good idea but this is forcing it to happen all at once without any additional resources.
Unfortunately, this is not the case. You can detect only the most obvious cases of the output from these tools. The distinctive presentation of these tools is a very intentional design choice - partly by the construction of the RLHF process, partly through the incentives given to and selection of human feedback agents, and in the case of Claude, partly through direct steering through SA (sparse autoencoder activation manipulation). This is done for mostly obvious reasons: it's inoffensive, 'seems' to be truth-y and informative (qualities selected for in the RLHF process), and doesn't ask much of the user. The models are also steered to avoid having a clear 'point of view', agenda, point-to-make, and on on, characteristics which tend to identify a human writer. They are steered away from highly persuasive behaviour, although there is evidence that they are extremely effective at writing this way (https://www.anthropic.com/news/measuring-model-persuasivenes...). The same arguments apply to spelling and grammar errors, and so on. These are design choices for public facing, commercial products with no particular audience.
An AI detector may be able to identify that a text has some of these properties in cases where they are exceptionally obvious, but fails in the general case. Worse still, students will begin to naturally write like these tools because they are continually exposed to text produced by them!
You can easily get an LLM to produce text in a variety of styles, some which are dissimilar to normal human writing entirely, such as unique ones which are the amalgamation of many different and discordant styles. You can get the models to produce highly coherent text which is indistinguishable from that of any individual person with any particular agenda and tone of voice that you want. You can get the models to produce text with varying cadence, with incredible cleverness of diction and structure, with intermittent errors and backtracking and _anything else you can imagine. It's not super easy to get the commercial products to do this, but trivial to get an open source model to behave this way. So you can guarantee that there are a million open source solutions for students and working professionals that will pop up to produce 'undetectable' AI output. This battle is lost, and there is no closing pandora's box. My earlier point about students slowly adopting the style of the commercial LLMs really frightens me in particular, because it is a shallow, pointless way of writing which demands little to no interaction with the text, tends to be devoid of questions or rhetorical devices, and in my opinion, makes us worse at thinking.
We need to search for new solutions and new approaches for education.
Thank you for that and for everything you wrote above it. I completely agree, and you put it much better than I could have.
I teach at a university in Japan. We started struggling with such issues in 2017, soon after Google Translate suddenly got better and nonnative writers became able to use it to produce okay writing in English or another second language. Discussions about how to respond continued among educators—with no consensus being reached—until the release of ChatGPT, which kicked the problem into overdrive. As you say, new approaches to education are absolutely necessary, but finding them and getting stakeholders to agree to them is proving to be very, very difficult.
You have to understand, you are a smart professional individual who will try to avoid being detected, but 6-12th grade students can be incredibly lazy and procrastinate. You may take the time to add a tone, style and cadence to your prompt but many students do not. They can be so bad you find the "As an AI assistant..." line in their submitted work. We have about 11% of assignments are blatantly using AI, and after manual review of over 3,000 submitted assignments GPTZero is quite capable and had very few (<20) false positives.
Do you want teachers wasting time loading, reviewing and ultimately commenting on clear AI slop? No you do not, they have very little time as is and that time will be better spent helping other students.
Of course, you need a process to deal with false positives, the same way we had one for our plagiarism detector. We had to make decisions many years ago about what percentage of false positives is okay, and what the process looks like when it's wrong.
Put simply, the end goal isn't to catch everyone, it's to catch the worst offenders such that your staff don't get worn down, and your students get a better education.
You could ask the student to start wrkting on google docs, and whenever someone gets a false positive, they can prove they wrote it through that.
And Besides 99% of people who use AI to write, dont bother claiming it as a false positive, so giving students the right to contest that claim would not be that much if a problem long term.
We let them resubmit a new paper when they are caught, and they get some one on one time with a tutor to help move them forward. Typically they were stuck or rushing, which is why they dumped a whole AI slop assignment into our LMS.
I'm also surprised that academics rely on snake oil software to deal with the issue.
Instead, academics should unite and push for outlawing "AI" or make it difficult to get like cigarettes. Sometimes politicians still listen to academics.
It is probably not going to happen though since the level of political apathy among academics is unprecedented. Everyone is just following orders.
Looking at some of the most successful historical pushbacks against technology, taxes and compensation for displaced workers is about as much as we can expect.
Even trying to put restrictions on AI is going to be very practically challenging. But I think the most basic of restrictions like mandating watermarks or tracing material of some kind in it might be possible and really that might do a lot to mitigate the worst problems.
Watermarking output (anything that is detectable that is part of the structure of the text, visual--if even imperceptible--image, or otherwise integrated into whatever the primary output is) will make it take a bit more effort to conceal use, but people and tooling will adapt to it very quickly. Tracing material distinct from watermarking -- i.e., accompanying metadata that can be stripped without any impact to the text, image, or whatever else is the primary output -- will do the same, but be even easier to strip, and so have less impact.
this is also not a good era for politicians to listen to academics, anti elitism sentiment is at a high and nobody will vote for "eating their vegetables" vs. "candy for dinner".
Prohibition does not solve the problem of needing to detect violations of the prohibition.
> or make it difficult to get like cigarettes.
Cigarettes aren't, at all, difficult to get, they are just heavily taxed.
I’m selfishly so glad I dodged this particular bullet in my studies.
Choose a randomly selected question.
Record and transcribe.
No need for AI detector.
_Maybe_ you can detect default ChatGPT-3.5 responses. But if a student does a bit of mucking around with fine-tunes on local llama or uses a less-common public model, can you still tell?
I have a similar question for AI art detectors. Can it actually work? Maybe it works for Midjourney or whatever, but the parameter space of both hand-drawn (on a computer) art and brush-stroke generating models like NeuBE must overlap enough that you could never be sure in a substantial number of cases.
We are seeing this with Grammarly already, where instead of a nuance Grammarly picks the beige alternative. The forerunner was the Plain English Campaign, which succeeded in official documents publicised in imprecise language at primary school reading level, it's awful.
- To the AI detector: "update your AI detection based on this new set of AI-generated content"
- To the AI writer: "update your AI writing to evade this new AI detector"
- Repeat
This is keystone cops.
Imagine two scenarios: five years ago somebody saw this coming, and they thought we should legislate a certain mechanism to prevent students from using AI assistants to cheat. Would we have done it back then? The answer is "no", since the problem was nebulous and we deal with situations like this after they come up, not before.
Now imagine a second scenario: somebody today tells you that AIs are on their way to own and supplant our societies. They are already functionally equivalent to regular human beings in many axes, and are only gonna get better at that. And thus, we should bolster our social apparatus with pro-human shielding... What do you say, should we deal with this problem after it comes up?
1. Would you give that power to a World Government to measure you in a behavioral scoring system hence technologically enabling bureaucrats to vote that error rate value somewhat as they do for interest rates today "to ensure progress" (uniparty propaganda)?
2. What makes that impossible to happen?