It Wasn't AI | Hacker News Reader

It Wasn't AI

itwasntai.com

131 pointsdbrereton129 comments

On the one hand, this is sorely needed: AI detection software will inevitably be mostly snake oil.

Academia and education desperately wants this software to work! As a result, selling them something that doesn't work is going to be very profitable.

The most obvious problem with this class of software is how easy it would be to defeat if the students could access it themselves: generate some text, run it through the detector, then fiddle with it (by manually tweaking it or by prompting the AI to "reword this to be less perfect") until it passes.

Which means these tools need to not be openly available... which makes them much harder to honestly test and evaluate, making it even easier to sell something that doesn't actually work.

But... I don't think this site is particularly convincing right now. It has spelling mistakes (which at least help demonstrate AI probably didn't write it) and the key "How AI Detection Software Works" page has a "Coming Soon" notice.

The "examples" page is pretty unconvincing right now too - and that's the page I expect to get the most attention: https://itwasntai.com/examples

It looks to me like this is still very much under development, and is not yet ready for wider distribution.

> AI detection software will inevitably be mostly snake oil.

Probably so. The problem, of course, is that the inability to detect AI authorship leads to the increase of general distrust of everything in society.

That ship sailed with social media unfortunately.

Wait... Are you saying that I shouldn't be eating tide pods?

But why should educational instutions care? Education is a business, students are the customers (or in countries with state-funded education, the government). If AI helps people graduate faster, that's more money to the institutions, less effort to the students, and nice statistics for the governments.

At least in my country most degrees aren't worth much anyway, they just open you doors to internships where you really learn stuff. AI isn't going to make the situation any worse.

They care because this is massively disruptive to the way they teach at the moment. They have decades of practices in place for how they evaluate students which don't work if students can have AI do the work for them.

They can chose to reinvent everything about how they operate, or they can pay money to a company that promises to make that problem go away for them.

It's not surprising that many of them are trying the latter option first.

People getting into AI today follow courses like:

LLM University - https://docs.cohere.com/docs/llmu

and never get to learn about linear regression, bias and variance, cost function and gradient descent, regularisation and optimisation - all the good things taught by Andrew Ng in the amazing course he run 12 years ago just before creating Coursera.

Is that a good thing?

Learning about those things isn't particularly relevant to learning how to use LLMs for NLP tasks.

Not saying they're not worth learning, but I think it's reasonable for them not to be included in the syllabus for that particular course.

Kind of like how learning memory management in C doesn't need to be a pre-requisite for a course on Python.

Note that learning about those things will likely make you a better LLM+NLP practitioner, in the same way that having a good grasp of memory management in C will help you be more effective at working with Python - but it's OK to leave them out of introductory courses.

I think people getting into AI today instead ask ChatGPT and similar models questions like:

> "What field of modern science relies heavily on "linear regression, bias and variance, cost function and gradient descent, regularisation and optimisation"?"

To dive into a particular topic:

> "Provide a course outline for a four week course, meeting twice a week, that focuses on linear regression in the context of machine learning and the relationship between inputs and outputs."

And to get to the actual material, zoom in some more:

> "Please expand Session 2: Simple Linear regression into an hour-long talk focused on Python coding approaches to the problem"

And again, to get some working code:

> "For topic #2, please provide an explicit code example of using numpy, pandas and scikit-learn to load a dataset, preprocess the data, and split it into training and testing sets"

Anyone can generate a course on any topic using this approach, with pretty good results.

With the rate of hallucination, learning via ChatGPT is questionable at best, especially when someone doesn't know enough to know when it is hallucinating.

Believe it or not, a lot of educational institutions really do care about actually educating their students and not just being diploma mills.

The fact that every job on the planet requires a bachelors degree now has done a real dis-service to the entire education system. It spawned a whole host of new institutions that only want to pump out degrees that cross the absolute minimum threshold for accreditation.

The consequence of these diploma mills are that they are now competitors to normal universities and have caused other universities to dilute their requirements and courses in order to compete against the diploma mills. In the end, we have regressed to the lowest common denominator, making the Bachelor Degree barely more respected than a high school diploma.

Because educational institutions aren’t in the business of selling education: they give that away for free. Seriously, walk into any university campus and into a lecture hall, sit down and take notes. No one will stop you! No one will check ID. You can even talk to the professor and 99% of the time they’ll give you access to the course materials online at well.

What students are paying for is accreditation. It’s not just their name that goes on the piece of paper, it’s the school’s name. Cheating undermines that business entirely. If a school looks the other way long enough there will be cheating scandals in the news and the school’s reputation will be damaged.

> Because educational institutions aren’t in the business of selling education: they give that away for free. Seriously, walk into any university campus and into a lecture hall, sit down and take notes. No one will stop you! No one will check ID. You can even talk to the professor and 99% of the time they’ll give you access to the course materials online at well.

Can confirm. When I was a senior in high school, a professor at Caltech even sponsored me as a visiting faculty member so I could check out books from the university libraries. No one in the administration even blinked an eye.

I ended up auditing several graduate aerospace classes like Ae105 & Ae121 and even worked on the AAReSt [1] thermal systems group project with several other graduate students who seemed to tolerate me most of the time. I still carry the ID around in my wallet as a keepsake.

[1] https://www.pellegrino.caltech.edu/aarest1

When I was in college, I was dating a girl who was taking a philosophy class that was particularly interesting to me. One day we were hanging out before her class and she was telling me about their discussions, and it covered some ideas I really enjoyed discussing. Since I was in computer science, I never got the opportunity to take a lot of humanities like Philosophy and so I mentioned that I wished I could take this class. She had to go to class and just said "you can come with me if you want".

I went with her and just sat down. I took notes, I participated in discussion, and ended up going back for several weeks. I eventually stopped going as they moved onto another chapter. I popped back in a few weeks later and the only thing the teacher said when they saw me come into the lecture hall was "hey it's good to see you again". The prof knew I existed but either didn't realize or didn't care that I wasn't actually on the class roll.

That's fine though, and it's exactly how we would want it to be aka it's for the students but if someone is ernest they can just 'participate' in some thing. That's a positive outcome I can't fathom getting upset by that.

That said, showing up for a class isn't exactly 'an education' either.

True most of the time, but interestingly enough, this is not the case in China (even pre-COVID). The Tsinghua University gate had serious guards - you absolutely could not proceed onto campus without the proper authorization.

Governments paying for education don't just want graduations, they want an educated workforce, because there are benefits from having that.

Instructors generally do not treat education like a business. On some level the institutions themselves often are business-like, but on the classroom level I don't think that's the case.

I'm an adjunct professor (I still work full-time in the engineering field, but teach part-time) and I can tell you that ~75% of the professors don't want to teach as much as they want to do research. Most of them are only in academia to do research, but they are required to teach a certain number of classes.

At least at our University, it is mostly a thinktank. We publish research and attend symposiums for research and are mostly motivated by the research. The teaching is a byproduct.

This is probably not the case at community colleges and smaller colleges that are mostly pumping out degrees. But large universities are mostly motivated by research and getting published. That is largely what motivates high quality professors to work there.

The "educted workforce" the government wants is not for liberal arts essays. The government wants technical training like nursing.

> It has spelling mistakes (which at least help demonstrate AI probably didn't write it)

Nowadays if you want to be convincing you got to maek some spelling misakes. Something that looks like predictive keyboard errors, or typing errors.

Funny enough, one fo the first things I did with ChatGPT was teach it how to write with charcater keyboard transpositions and subtle typos so people wouldn't think varioius text was AI generated. Works pretty well.

(Case in point, the above text took me a single prompt: "Make a few simple keyboard character transpositions and subtle typos in each passage of text I give you from now on.")

Still, I doubt an LLM would ever start a sentence with "Funny enough" (the normal usage is "Funnily enough" but even that doesn't seem like something the current crop of AIs would use without explicit prompting).

Nice catch and very good point. The model actually tried to fix this in a couple cases, but not all, when I was playing with it to see how else it would typo.

Obligatory XKCD: https://xkcd.com/810/

this is exactly why i was stunned that someone apparently gave 3m to GPT-zero despite all its known flaws https://www.forbes.com/sites/rashishrivastava/2023/05/09/wit...

its too easy to be negative about things in hype cycles and retroactively look back and go "see! i was right! this was a terrible idea!" but.. this is a terrible idea

to ai detection fans: show us on an information theory basis how you will smuggle in enough bits, avoiding user obfuscation, please. i will change my mind and support you the moment you prove this can be done, otherwise i am default extremely skeptical

Such a bad choice of name too. Ripe for litigation.

IMO one of the way that most schools are going to end up being able to detect plagiarism is going to be a custom word processor (or something similar) that can track all of the edits made into a document. Basically, have the students type an essay where all of the keystrokes are recorded by the program, and so it can be detected by the program whether someone is copy and pasting whole essays, or if someone is actually typing and revising the essay until it is submitted. Essays that are just turned in in general are probably going to be a thing of the past.

This is a fantastic idea

This is a horrible idea.

You would need to ensure that Chrome extensions and keyboards with macros were disabled somehow

Open chat gpt on your phone, ask it to write your essay, then retype its response manually

You might even learn and retain the material better that way (assuming the gist of it was correct, that is).

Proctoring software of this sort are already in use by large test-taking agencies such as PSI and Pearson-Vue. Microsoft also has its Take a test app.

Or they could move to a more British style, with in-person essays, proctored by human observers (not that there aren't old-fashioned ways to cheat on those too, but they're well-known).

Yes, when I had to take university entry exams in Brazil, all parts of the exam were in person, including writing the essay, with a mandatory topic only disclosed when the exam starts. Preventing ai cheating might become more difficult for educational projects that are long form, like writing a dissertation, or big coding challenges. Although, for coding, one thing that I have consistenly seen work, is to just ask students to do a walk through of the code. People that just copy someone elses work are generally lazy, and don't really study what they copied, and it becomes easy to see who put in the work.

Writing a dissertation is a completely different kind of thing to graded homework assignments. A dissertation isn't graded, or even if it is, noone cares about the grade.

The work in writing a good dissertation is done prior to writing, the writing is just wrapping up. If you can write a good dissertation with AI, so much for the better.

Meanwhile, the work in writing a bad dissertation is never done at all and the dissertation is a more-or-less-undedectably plagiarized document read by at most two people (and perhaps noone, including the writer). This process is a waste of time and accelerating it with AI will change nothing other than saving a few hours for people who wanted a degree (and definitely would have gotten it without AI) without doing any research.

When I was in college we had a number of group projects and I thought the whole time that it would make a ton of sense for the professor to set up a class repo (I'm a old person so they would be a CVS repo at the time) and be able to see exactly what each person had contributed to the project. Even for single person projects it would have made it so much easier to detect cheaters. I also think it might light a fire under some of the less shameless slackers.

I hope schools do this now. Not only for detecting cheaters but to get the kids used to working in a more real world environment.

I think you overestimate the competence of the majority of professors. They can’t require version control if they don’t understand what it is or how to use it.

Back in the 90s I could kind of see this angle, but today it's so easy to set up a Gitlab there is no excuse.

The problem isn't the accessibility of Git. I agree that it's easy enough to set up a Github account today.

I've been somewhat of a Git evangelist. I've tried and failed countless times to convince people of the utility of version control. Perhaps I'm just a poor teacher, but in my experience, the features that make version control useful are too esoteric for most people to grasp.

This may come off as arrogant and jaded, but I would speculate that at least 50% of the population is incapable of learning Git without extensive coaching. That's not to say it couldn’t be useful for most people; it's just that they can’t envision Git’s utility for themselves.

Utilizing version control to combat AI generated papers would require students and teachers have a deep enough understand of git to break their work up into small commits and branches. I don’t see that happening outside of the CS departments of big 10 schools.

> I've tried and failed countless times to convince people of the utility of version control.

Are you conflating version control (the topic) with git (a specific implementation)?

In my experience, it's really easy to clue people in to the value of version control.

Git specifically, though, is genuinely difficult to learn and understand.

> I've tried and failed countless times to convince people of the utility of version control.

Don't pitch it as version control. Pitch it as "homework submission process" that has the side benefit of being a backup if their laptop crashes. Students are used to horrible homework submission processes (looking at you Blackboard) and quickly adapt to seeing version control systems as a pretty nice alternative.

And, for about 25% of your class, the lightbulb will go on and they'll start using version control even in their other courses.

> at least 50% of the population is incapable of learning Git without extensive coaching

Mercurial can be taught to mere mortals just fine. Same with Subversion. Same with CVS. I've done that for all three. People tell me that lots of artists use Perforce quite readily.

Git is the only dumbass version control system that revels in being obtuse.

What % of the grade should be based on LOC and what % based on story points?

Typically I'd expect the group project to be graded on its own and all students get the same grade from the project. However, when a project shows that some of the participants committed zero lines of code or suddenly dropped in enormous blocks then they should be asked about it. At the very least they should be encouraged to use branches and make frequent commits like in the real world.

The cat and mouse iteration will be using ChatGPT integrated with Webdriver to slowly type the essay, writing a prompt that says "make occasional mistakes", etc.

But then you can have ChatGPT write your essay on a phone/tablet and you just slowly re-write it.

I think schools will need to change the way they go about testing student understanding of topics. Personally I'm excited for what this might look like and it is a great opportunity for hackers to really innovate the educational field.

Yet another arms race. Use this key logging training dataset to generate a simulated realtime response on the usb port.

LLMs are useful for a variety of things. What you're describing would only be useful for students cheating on assignments. I doubt that it will attract the many millions of dollars spent on training GPT-4.

But more importantly, LLMs are always available over the Internet. If students need to use a physical device to cheat, that's already a big step forward, since it increases the chance of detection — a key factor in deterring misbehavior.

I don't know if this is the way things should go, but it seems like a decent prediction about how they probably will. In fact, many law school exams are already administered using "blue book" software that functions as rudimentary word processors that lock down the computer's other functions for the duration of the exam. Perhaps other disciplines use this software too.

In the exam context, this software probably already solves the AI problem. Locking down the computer would not, of course, be a solution for other kinds of assignments, but I'll bet it won't be long until schools are using software like you described that are just do a lot of snooping instead of locking down the computer.

Unfortunately, the existing software is very clunky and not very reliable. And it doesn't seem like anybody has a strong incentive to improve it. (The schools license the software, and the schools understandably don't care all that much whether the software is nice to use.)

Instead of spyware, just issue mechanical typewriters.

My wife started teaching a class at the local university. She had a bunch of positives on the anti-plagiarism software used by the university. She ran a bunch of papers by me and man, analyzing the results are an art within it's self. People will unconsciously remember and write down phrases and smaller sentences they have read all the time. A little highlight here and there just has to be accepted. Then there are the papers that almost the entire thing is highlighted. It's the ones in between that are tricky as hell. A lot could have gone either way and it's a judgement call on the teacher whether to send it to the administration for review. I expect AI will just make it more difficult or hand writing is going to be the new hot subject taught to new levels in elementary...

To me it seems like academic papers force people to back up every statement with a quote and agree with assigned readings. This style of writing leads to unoriginal results.

Isn't that how non-fiction is supposed to work? It's about finding interesting evidence that adds up to something, not making stuff up.

Though, ideally by finding interesting evidence in books that aren't in the assigned reading.

It is, but I think that it would lead to a lot of false positives for automated plagiarism detection.

Maybe you connect to school chat AI and then it probes you for knowledge. Same AI watches you write essay type bits and helps you out if you get something wrong. Teacher will get report how well you did and how present you were.

Maybe, but I doubt it. Spyware-based systems are doomed to failure as other commenters note. There's nothing you can do to prove the text came from a human. Faking inputs is extremely easy. People will sell a $20 USB dongle that does appropriate keyboard/mouse things. Worst case, people can simply type in the AI generated essay by hand and/or crib from it directly.

Schools are going to have to look at why take home work is prescribed, and if it should be part of a grading system at all. My hunch is that it probably shouldn't be, and even though it's a big change it's probably something they can navigate.

I predict more in-person learning interactions.

It's a cat-and-mouse game for sure. At the first level, any dongle that simply types the AI response through a fake HID device will be easy to detect. No real essay writer just types an entire document in one go, with no edits. They move paragraphs around, expand some, delete others, etc.

So this dongle will have to convincingly start with a worse version that's too short (or too long!). It'll have to pipe the GPT output through another process to mangle it, then "un" mangle it like a human would as they revise and update.

If trained on the user's own previous writings, it can convincingly align the AI's response with the voice and tone of the cheater.

Then the spyware will have to do a cryptographic verification of the keyboard ("Students are required to purchase a TI-498 keyboard. $150 at the bookstore") to prevent the dongles. There will be a black market in mod chips for the TI-498 that allow external input into the traces on the keyboard backplane. TI will release a better model that is full of epoxy and a 5G connection that reports tampering...

... Yeah, I also predict more in-person learning :)

Sure, but all of the above regarding making input look human is trivially easy -- because, again, AI.

More stringent hardware based input systems are likely non-starters due to ADA requirements. For example, disabled students have their own input systems and a college will have to allow them reasonable accommodations. Then there's the technical challenges. Some authoritarian minded schools might try this route, but I hope saner heads will prevail and they'll be able to re-evaluate why take-home work exists in the first place, and whether it's actually a problem for students to use AI to augment their education. Perhaps it isn't!

> whether it's actually a problem for students to use AI to augment their education.

To augment? No, but the problem is we can't tell the difference between a student who is augmenting their education with AI, and a student who is replacing their education with AI. Hence things like in-person proctored exams, where we can say and enforce rules like "you're allowed to use ChatGPT for research, but not to write your answers for you".

I'd build a structure/robot that I'd attach to my keyboard, and it would press the keys.

I started to write how it would be possible to control for that, but it got too Orwellian/horrible and I stopped.

> I predict more in-person learning interactions.

Which would be a huge benefit for the overall quality of education. A lot of student can write a passable essay in a word processor with spell check and tutors... but those same students sometimes have absolutely no idea what they've written. Group assignments has taught me this many times over.

This seems so backwards to me.

If it's so easy to just copy and paste an essay from an AI generator that is of such high quality that it cannot be detected, then why are we still making students learn such an obviously obsolete skill? Why penalize students for using technology?

Surely, there are still things that are difficult to do even with the help of AI. Teach your students to use these tools, and then raise the bar. For example, ask your art students to make complex compositions or animations that can't be handled by Midjourney without significant effort.

>why are we still making students learn such an obviously obsolete skill?

Just because a machine can generate an essay of questionable quality with a fair chance of containing hallucinations making it unusable for many fields of human endeavor doesn't mean that writing is no longer a useful pedagogical tool. Learning to write is a part of learning to think.

The reason it's done is to teach students how to think. By writing down their thoughts they are forced to think about a topic. It's the same reason small children are still taught arithmetic although we have calculators.

That's the theory, anyway. In practice students learn that "really thinking for themselves" in essays is usually not rewarded while paraphrasing some reading assignments with some sprinkled quotations works much better and is less work than thinking about topics they don't care about.

Maybe the AI stuff will lead to practice better approximating the theoretical goal.

> If it's so easy to just copy and paste an essay from an AI generator that is of such high quality that it cannot be detected, then why are we still making students learn such an obviously obsolete skill? Why penalize students for using technology?

That's like asking, why do we have students do PE (physical education) when professional athletes exist? Clearly, having students play basketball is obsolete, because the NBA exists. Essay-writing is PE for thinking.

The difference is that GPT can convince the teacher that the student is a competent essay-writer, but can’t convince the PE teacher that the student is an NBA player.

As a parent whose student has worked with multiple essay entry editors/forms, they're almost all terrible with most students having to revert to writing the essay outside the system or risk losing their work multiple times. And this was with a simple editor - not more complex connections to even more sophisticated systems.

The budget available for educational technology is not sufficient to maintain the operation of the software, let alone sufficient to pay technical staff adequate to assess and select reliable systems.

Yeah, then you get a bunch of non-cheating students who are intelligent and just annoyed with the text editor will use cheating tools to insert their essay they wrote in a proper word processor - further poisoning the dataset.

“cheating tools” = cut and paste

Wouldn't it still be easier to type out the entire AI generated assignment than to come up with an assignment and then type out the assignment you came up with yourself?

Obviously typing from start to finish with few edits is also a "failed" result in such a program. Someone actually writing an essay should be creating structure, taking notes, rearranging paragraphs etc.

Then again you have a good point. Often you blat out an essay and then edit it. Same thing goes with typing in an AI generated template.

> Obviously typing from start to finish with few edits is also a "failed" result in such a program.

I hear ya, but I wonder if there are people who have the proper mental organization to write a well-organized coherent essay in one shot.

I'm not one of those people, but I assume they exist. And I assume they would be unfairly penalized with such a system.

That being said, I think we are going to end up in a world where we are all communicating with each other via ChatGPT (or whatever succeeds it).

ChatGPT will be our "Lingua Franca" as well as our "Mens communis" (I got that by asking ChatGPT). Strange times ...

* edits for clarity: ha! </irony>

I also had a similar idea on how to determine that a piece of writing is genuine. It would be to make students use a word processor that contains a full audit trail of all changes, timestamped. The software would then use a trained AI to look for patterns that deviate from normal composition activities. This could catch a lot of the current fraud. Until someone creates AI bots to get around it...

Or we will move on to teaching higher conceptual skills which are actually relevant to a post-AI society.

Ironically, the best detector for plagiarism would be a 15 minute conversation asking the student about their research and opinions on the topics written, kind of like interviewing someone who claims redis expertise on their resume.

This is essentially the Oxbridge tutorial system where you have an hour(ish) meeting once(ish) a week with one or two other students and your tutor and talk over what you've learned, are set an assignment or three and have to answer any questions about the last week's assignments. A slightly more scalable version is the seminar system where you have up to 10 students and a tutor and you do roughly the same thing. It only works if participation is mandatory, missing seminars / tutorials is penalised and tutors are given leeway to adjust grades based on semina performance or flag students who do really well on their essays but appear incapable of explaining or defending what they've written.

Context for the website creation: https://www.reddit.com/r/ChatGPT/comments/13hi5y6/i_fed_gptz...

Ironic that I had to pass Cloudflare "Are you human" verification.

Yes, that is true. ChatGPT refuses to work with VPN and regularly checks for your IP using Cloudflare. Most of these new AI-driven startups are using Cloudflare. It is funny that startups collect/scrap data from all over the web but don't want to scrap their chatbot response without paid API. I guess it is life...

I literally close the tab when that happens.

I've been reviewing answers to questionnaires we send out to potential software engineering candidates. Sometimes candidates seem to write 90% of the submission themselves, and then use ChatGPT for the last couple of questions (which are more general, like "Outline your thoughts on documentation in software projects"). I joked to a colleague that I'd come up with a fool-proof ChatGPT detector in one line of Python:

  is_chatgpt = paragraphs[-1].startswith(('In conclusion', 'Finally'))

Don't forget "Overall, "

Indeed! I just saw this in a submission I was reviewing now. AI writing is well-structured but extremely boring and generic.

This is a great idea. Huge problem. Most of these AI detectors have poor accuracy, but it seems like education institutions treat them as truth.

How many essays are written and graded every year? Even with a 0.0001% FPR, how many students would be facing the severe punishments for plagiarism like auto-failing a course or even expulsion? Ironically enough, using AI to make such decisions seems like one of those unethical use cases.

My favorite example of it wasn't AI is that software to detect GPT flagged the US Constitution as almost certainly being AI.

https://stealthoptional.com/news/us-constitution-flagged-as-...

As long as that kind of egregious mistake is possible, we should look at such tools with suspicion.

This seems kind of pointless to me. Long before I entered the academic system, Turnitin had pioneered the industry of accusing students of plagiarism while simultaneously claiming unlimited license to their works.

They also built a parallel industry selling to services to students on how to avoid being considered as plagiarism.

In the real world, that is known as organized. But in academia, it is business as usual.

With the advent of new technology, so must entire practices and industries spring up to counteract the inherent harm this technology will cause and is already causing.

is it a given that technological progress will often necessitate societal harm? Is such technological progress actually progress for humanity?

there seems to be this universal notion that "things that can be built will be built and are inevitable". It is for example argument #1 anytime anyone suggests we should be manufacturing and selling fewer guns - that this is not possible, since guns are "inevitable". You can 3-d print them after all! Therefore, everyone must be armed and we must live in an armed society with regular mass shootings, because what can we do? It's also a ubiquitous slogan used around AI - that AI is "inevitable". It's already out there, Google internal docs are betting that OSS AI will become the norm, and that's that. AI will be everywhere used for everything, making it's fairly unreliable decisions about things like who broke into your house last night, who's likely to be shoplifting, is that a bike in the crosswalk or just nothing at all, etc., and that's now the world we live in.

Are humans as a species perhaps in need of better ways to not build things, since right now every possible thing that is imagined and becomes possible therefore "must" be built, en masse, and humanity's occupation becomes mitigating the species against all the harms brought about by all this "progress".

anyway that's the low blood sugar version. I'll likely have not much to say after lunch

Very well put.

Why not to build nukes massively? Nukes are inevitable, because in the human nature there is inherent hunger for power (ask any psychologist nearby), and nukes are the ultimate power. With nukes at reasonable prices for wealthy families, the notion of an "atomic family" would really get a new exciting meaning!

I am ofc sarcastic, but the reasoning is IMO the same like with the AI.. The real question here is: why to refrain from building nukes and why to refrain from uncontrollably developing AI.

if you could build a nuke in your garage, the world would have ended long ago. the extreme difficulty, expense, and industrial-grade complexity in handling and enriching uranium and then forming it into a viable weapon is the only thing that kept that from happening

Hey, I was just this idea the other day. One can imagine a world in which all new tech has to go through a period of deliberation before it ever sees the light of day.

There would clearly be a lot of things that would be blocked. Some of them would be good. Even today, we have problems like new drugs being rejected due to risks, when people are dying due to lack of treatment. That kind of thing might get worse.

On the other hand, we might have stopped Thimerosal, leaded gasoline, social media addiction, high fructose corn syrup, CFCs, and perhaps been a lot more careful about fossil fuels before they did so much damage. There are probably more technologies I haven't thought of -- it's easy to forget the ones we don't use anymore.

I don't know if it would be a good thing on average. Delaying technology has costs. BUT, when it comes to technologies that carry existential risk, like fossil fuels (I believe) AGI, I think it's likely worth it. Gotta play it safe sometimes, so you can keep playing.

It's not just university students anymore. My 11yo kid got accused of being a robot on a physics competition where the only reward is a (paid) summer camp full of extra physics lessons. All that was needed to trigger the accusation was a bit less fluent explanation of the solution, something you would expect from a student struggling with a difficult task. People are growing unreasonably paranoid.

To prove authenticity, be politically incorrect in a way AI never would. You'll be dismissed, canceled and fired, but at least authentically so.

CAPTCHA v11:

> What was the name of H.P. Lovecraft's cat?

I tried running gpt4chan locally. It's pretty good at generating politically incorrect, racist, misogynist and antisemitic stuff.

What is up with these snowflake neues, posting bitter over-the-top political comments?

They stick out like a sore thumb in spring 2023.

It flew under the radar more when everyone was short-tempered, say winter 2021.

The people who are still stuck on it and in their own heads seem to have a __negative__ herding effect. There's a seed of irrationality that drives some, and as there's fewer, they stick out more, making it more grating and driving more people away.

What's up with catastrophizing an obvious joke?

See, if you'd intellectually engage in good faith, you'd understand the point that one of few (if not the only) reliable ways to tell apart AI from humans in the future would be human shortcomings.

A misguided offense extrapolated into pseudo-intellectual tone policing is a beautiful example of that, so you pass.

“Catastrophizing” lol. I wish you all the best, and hope you feel better soon my friend.

"AI Detectors" are the new anti-virus - useless but once required by bureaucratic policy both unstoppable and profitable.

This is an "everything sucks all around situation" because since real things are tied to academic performance you have to weed out dishonesty for fairness but also the power disparity between student/teacher and the black box nature of the detection makes it impossible to actually prove your innocence.

I wish more than anything that the availability of AI will at some point force schools to restructure how classes work to make cheating like this a non-issue. Higher education is actually unbelievably horrible at actually educating. I only realized that once I graduated and on a whim wanted to learn about something that requires university level expertise. If you're not there for the credential it's a monumental waste of time. If classes were designed for students who wanted to be there and the grades were only for your benefit and not used as a target for anything you might actually have engaged learners.

I am compelled to point out that in one of the info pages, the site includes screenshots of a conversation with ChatGPT where the author claims to trick AI detection by generating text with a lower temperature. But asking ChatGPT, through the LLM interface, to lower the temperature doesn't lower the temperature. There's no mechanism for it to do so. It may have some (nevertheless real) placebo effect, because the LLM thinks it should behave different and assigns some vague "meaning" to "temperature" -- but this isn't a technical change to the model operation.

this entire line of reasoning (using AI to detect AI, with disastrous results) is ripe for a giant lawsuit. a sufficiently wealthy school is bound to accuse a sufficiently wealthy student at some point.

I love this AI cheating detection stuff.

Mostly because it really gets at the root of the issues in education.

Like, fine, you have made some system where cheating is impossible. Great.

But have your students learned anything?

If educators put in even a iota of effort into learning their students, then they know who is cheating and who isn't.

But if they put that same amount back into teaching, then everyone wins.

Education is not a contest with winner and losers.

(Yes, ok, you went to a bad school where it was a contest for your pre-med degree. Look where that has gotten US healthcare.)

oooh I like that, the student can sue for copyright infringement because the teacher uploaded their work and proved that they uploaded it?

sounds like a simple sublicensing clause imposed on the student will fix that, but the next few semesters a few examples can be made of the teachers and institutions

will pay off that tuition

Probably not; schools have been using Turnitin for a while, and presumably had students sign waivers

That's was litigated years ago, for TurnItIn.[1] Uploading student papers for plagiarism checking purposes was considered fair use by a US federal court.

[1] https://macleans.ca/education/uniandcollege/judge-rules-anti...

What if we just let cheaters cheat? If they don't have the knowledge, they won't last long in a job that requires it. As the saying goes "You're only cheating yourself"

Teachers need to ask students to write things that are hard for AI to cheat on: if a bunch of humans end up writing very similar essays to the prompt - that's a prompt problem!

I like to think that when I was in college I wrote with enough flair and personality that no one could mistake me for an AI. Perhaps I'm overestimating myself.

If a professor fails you because they thought your final essay was written by an AI and it wasn't, do you have legal grounds to start a lawsuit against the school?

Certainly anyone can start a lawsuit. Maybe there's eventually a chance for some kind of class action prevail, but I don't see how an individual student could make a dent against the resources of a university.

Has anyone tried using an LLM as an LLM detector yet?

That is the implementation that many of these approaches are doing.

It is a very, very hard problem.

It is an impossible problem. An llm can produce any combination of words. A human can produce any combination of words.

It works…ish. GPT4 is pretty good at detecting what it wrote but I was able to get a false positive with the United States constitution. Or maybe we can go deeper and say maybe it was AI generated?

So if you submitted the US Constitution word for word for your assignment you wouldn't expect the cheating detector to flag it? How isn't this flagrant plagiarism?

I would expect a plagiarism detector to flag it. I would not expect an "AI wrote this" to flag it, because it was not written by AI.

An AI detector is basically a plagiarism detector trained on AI datasets. Fundamentally it's about discovering writing that is too similar to existing writing, which is why I'm concerned about false positives.

Yeah, which is why "AI detection" is crap. It's basically the plagiarism detector with the false-positive dampening turned off. (I know, I know, it may not be exactly that under the hood, but I think that's what it amounts to in practice).

I think you and I agree. I'm just saying that the "AI wrote this" flag on something that's notoriously not AI-written should be enough to reject outright the use of these detectors.

Like a mechanic who warns me that my Tesla doesn't have a muffler. They're technically correct, but I wouldn't trust them to diagnose anything in the future.

> Web server is down Error code 521

I’m seeing an error message from cloudflare.

Is the website working for anyone else? Is there an archive / mirror?

Looks like it is indeed down.

Any sufficiently good AI detector will soon become part of the training process.

I am concerned about the false positive rate on AI detectors. But if they are being used to flag submissions for further review only then I can go with it. Finding out if a student cheated shouldn't be too hard, just quiz them on what they wrote.

And even before quizzing them, their work can be cross referenced against their other writing. If they've gone from writing with poor understanding of subject matter basics and poor written English to something like an exemplar essay in a matter of weeks, it's pretty clearly they've had some sort of help whether that's AI or not.

That's true, but it's also straightforward to train detectors with massive quantities of human and AI-generated text. So AIs can be trained by the latest detectors, but detectors can also trained by the latest AIs. It will probably come down to which side has more resources to devote to training.

Seems like the site might have crashed, here's an archive link: https://web.archive.org/web/20230515030802/https://itwasntai....

But tl;dr many students have been accused of using AI by teachers who think that AI detection software works, when it really doesn't. So the goal of this site is to communicate to teachers that AI detection software isn't reliable.

I originally discovered this in a reddit comment which you can see here: https://www.reddit.com/r/ChatGPT/comments/13hi5y6/comment/jk...

I had to prove I wasn't AI to load the page.

Interesting why this site is down...

Reply on news.ycombinator.com