My daughter is only 1 and a half years old now, but my concern is more that she'll be implicitly competing with classmates that are using tools like this, whether we allow her to use them or not.
The teachers set the course material and grading standards at least partially on how well the students are performing. Maybe not for a given class or year, but certainly over time. Scholarships are competitive. Slots in higher level courses are competitive, and often (at least partially) based on grades.
Can you imagine that the coursework and education overall might, over time, look quite different if half or more of students are regularly using LLMs, without explicitly disclosing it?
Something like that was advocated by the Khan academy guy, but I'm not sure if he worked out a full replacement system. There are some things in the current system like honors classes or retaking the classes for people who got an F, but why have the F ruin their chance at college if they later master it and get an A? If they always lag but eventually get there, I guess an argument is college would be too expensive if it took them a long time to get through it.
My father, who was a math teacher, already faced the problem mid-90s, as cheap mobile phones became available in my country. Things have only gotten worse since then. ChatGPT is only one more brick in the wall.
https://www.hubermanlab.com/episode/optimal-protocols-for-st...
I’m not doubting your claim but I don’t have time to listen to that podcast and I’m interested in what was said since you did.
I skimmed the linked material and that's not my conclusion. Practice tests are better teaching tools than just memorizing the material and then taking a real test, but on an absolute scale, I'm convinced that they're pretty terrible for most types of material.
Throughout my education I've come across countless people who could memorize the material and recite it with relative ease but they didn't have any intuitive understanding so they couldn't use that "knowledge" to solve real problems. Much like ChatGPT itself.
Equating good memorization and recall performance with good education and knowledge seems like a form of cargo culting to me, it's missing the essence of what makes knowledge powerful in the first place.
Regardless, I agree, attempting to assess the progress of a cheater is a fool's errand.
I would not wish to live in a society where “can bullshit an assignment with ChatGPT” is generally competitive in education.
The competition is very obvious in societies with a lot of wealth disparity. In Europe there tends to be less disparity and less competition.
US, China, India, UK, etc, much more disparity so it’s much more important for kids to be ‘top of the field’.
Fully agree.
> The competition is very obvious in societies with a lot of wealth disparity. In Europe there tends to be less disparity and less competition.
Fascinating observation! Wonder if that shows up at the school level as well? For example, In the US the highest/lowest performing schools tend to correlate to high/low income areas and have significant disparity in quality of education, and I expect this is a strongly self-reinforcing trend towards elite private institutions.
AP courses notwithstanding. AP courses were insanely competitive and speaking as someone had been all over the map in my K12 years (AP, accelerated courses, early foreign language, and a touch of remedial relative to grade level) the students in AP were mainly in competition with themselves. That however is just personal anecdote.
I also suspect that most kids, particularly grade schoolers, aren't aware they're competing in this sense. Parents are a different story.
Unless the child is a literal genius or very charismatic, wouldn’t they almost certainly have worse prospects after graduating in the 10th to 30th percentile of their class at Harvard than from the 80th to 90th percentile of their class at Boston U?
(Assuming they could even manage to squeak into Harvard in the first placd)
By definition to join a network they need to be accepted by their class peers.
But you'll be exposed to different people and opportunities by going to a state school instead of a community college, or a better school out of state with a scholarship vs. a state college, etc. The small differences matter.
I literally got my first job because my roommate's dad knew of someone looking for a software developer job. Would I have met that person in community college? Maybe! Lots of people with good networks go to community colleges, too - but I'd bet fewer than more expensive schools.
For example, if someone on HN says ‘a 50th percentile intern’ at a certain company, they probably wouldn’t be suggesting the intern literally has a gpa in the 50th percentile among all the other interns at that company.
For instance, having graduated from Harvard is going to get you a fairly good job _regardless_ of their GPA.
Similarly, there is no objective way to measure "merit" that isn't unfair to _someone_ and also scales to dozens/hundreds/thousands of students. So people use GPA.
Life isn't fair.
Clearly peers do evaluate each other in any ways they choose, pretty much every day at Harvard I imagine.
I would note that students who depend on AIs to answer questions on homework will do poorly on tests. I am not really sure what all the fuss is about. Kids cheating on homework is nothing new and a machine doing it for your is little different than all the other ways kids cheat on homework. And cheating on homework doesn’t help you in your grades - it hurts you because you’re unprepared for exams, which typically dominate the weighting of a grade. Then once you take a standardized placement exam you’re totally screwed.
How does this relate to the prior comment?
I’m clearly not an admissions officer at Harvard, nor likely are the other HN commentors here.
Parents are worried about the worst case scenario, and going to Harvard raises that floor by ensuring they can get noticed straight out of graduation, even if they are bottom of the class.
https://www.teamblind.com/post/twitters-internal-hiring-poli...
Oh wow, I thought Americans were super lax when it came to this, especially compared to more competitive educational systems like China. My kid is in second grade and hasn't gotten a real grade yet.
It's hard to generalize here, as Americans are a heterogeneous bunch. Demographic subsets exist where education is seen as a vigorous competitive sport, where the goal is to achieve admission to the "right" college. The parents are as involved as the kids, usually for better but sometimes for worse.
At the same time, among other groups of Americans, the goal is simply for kids to reach adulthood without too much interaction with the police. You can't draw sweeping conclusions that will cover the entire spectrum.
This will force ChatGPT-using classmates to either use it as learning _assist_ tool, to help student learn the material; or be 100% reliant on ChatGPT for homework, but fail every in-class assignment and then fail the whole class. Either outcome is fine with me.
It starts now for you by the way. Keep her away from screens for now, and be a good role model when handling devices (and the apps on them). You have ten years to show them what computers can do for you, but also what being the master of your own cognitive skills will gain you. If you do it right it won't matter whether she is allowed to use those tools or not, she'll use her own brain because that will allow her to get ahead by the time it really matters.
Learn to live on lentils and you will not have to be subservient to the king.
If you want your kids to be subservient to the king then of course they should use ChatGPT, and you should want schools to allow it otherwise we will just have more Asian and Indian students "getting ahead"
As a parent whose child has been in two school systems, respectfully: yes, it is. And it begins earlier than you think.
Everything has to be managed in moderation.
As for ChatGPT, I use it all the time for learning. And I've used it to help my kid study. But I wouldn't give my kid unfettered access to it just like I don't give him unfettered access to anything.
Sarcasm aside, making informed decisions on how to do those things can be a challenge.
That said, I hear stories like yours (from friends) and it always baffles me because that's just now how our family operates. I have a downtime pin on devices for the same reason I lock the door to my car even though it's like 50% breakable glass windows.
My kids know the rules. One of them for sure knows the pin. But they also know that if they use the device outside of our agreed upon guidelines, they're in trouble.
I'm sincere in asking, aren't behavior controls a lot better than technological ones?
Punishment and technological locks are both tools in the toolkit.
Ultimately you want your children to find the right balance with technology and that's difficult because most parents struggle with that balance. The world has moved very quickly and it is very hard to keep up. My kids are 12 years apart in age and they couldn't have had more different childhoods -- that's how quickly things have changed. Less than a generation.
My optimistic hope is that for the more basic skills, teachers can adjust grading so that more easily-cheated homework provides less credit, and in-person work (such as a pop quiz) is weighed more heavily. Then, you're effectively setting yourself a time bomb by using LLMs on homework to avoid learning the material.
For higher-level skills, I think using LLMs will probably just become another skillset and part of the toolbox, just like the Internet was for my generation. But I guess we'll see.
People act as if cheating on homework is new with LLms. It’s not. The homework is there to guide you towards the exam, and exams are generally proctored to prevent cheating in ways homework is explicitly not. Generally teachers grade homework to create an incentive for students to practice rigorously for the in course and final exams. Any student who doesn’t avail themselves of that practice invariably flunks their exams. Your homework grade won’t help you then.
So who cares? Sooner or later the cheating student either learns why it doesn’t pay to cheat or suffers the life consequences of not learning in school. This isn’t new with LLMs.
What is new is you have a tool that you can get advice from on topics that are difficult or even get up front grading and advice on homework before turning it in that helps you master the material better. Students who use this will be able to master the material faster, even if their parents both work and can’t invest the time to tutor their children at night.
(N.b., The threads here imply every parent works at a cushy tech job and has plenty of time to be invested in their kids education and any parent who isn’t is negligent - while the truth is many are just struggling to make ends meet - as my parents did … I didn’t get the investment my daughter gets not because they didn’t care but because they were busy feeding me and keeping us from being homeless. I would have loved to have ChatGPT help me as a kid)
Not in the US it isn't. Grades > Exams > Extracurriculars is the usual hierarchy.
Imagine the token weights etc to live an an N dimensional embedding.
Consider a random 2D isometric projection, now imagine plotting (a subset of) the tokens on screen at their respective locations, with the background having patterns conveying the orientation of the projection.
Imagine tokens appearing FIFO, and 95% or so of the tokens at their correct position, and 5% at incorrect positions.
The user is expected to identify misplaced tokens in the projection.
Each the frame the perspective slowly changes.
Since we guarantee ~95% of the tokens on screen are correctly placed this enables the human to absorb their correct high dimensional location subconsciously.
Imagine this game is made a bit addictive.
Imagine eventually the user gets high scores and thus knows most of the coordinates of the token embeddings in a coordinate free way (no axes had to be drawn)
At that point all the information the ANN required is undisputedly present in the humans brain, absent how those coefficients are used. This is where the fun starts.
The alcorithm can calculate the likelihood of observing 2 tokens in a certain order. It can generate more probable pairs, this will correlate with the information in the brain.
For example imagine the user knows some spanish, and that the language model was multilingual. Suddenly the brain starts picking up the correlation between close juxtapositions of certain tokens, and their positions learnt from the computer game.
While playing and getting better scores in the game, the user starts noticing its own grammar and vocabulary improve, because the brain has been helped in better estimating next tokens...
People will be able to learn math, languages, programmnig languages, tables of chemistry, etc... with substantially less effort.
So this kind of development adds to my pile of worries about their future, but for exactly the opposite reason.
The essay my 8 year old can currently write in scribbly handwriting certainly won't compete with the essay a classmate "writes" using ChatGPT. But the facade is paper thin and ever so short lived. Seeing how much better my kids write now, vs 2 years ago, proves how much they're learning and the incredible power of practice. Someone relying on such a serious crutch will be left behind. I guarantee it.
Unless there's some dystopian future where every aspect and communication in our lives are managed and filtered by some kind of AI middle layer. I doubt it.
Tech advancements always lead to this struggle, but it's rarely as stark as you might think. My kids are still learning cursive in school, though I can't imagine when they'll ever use it.
What makes this different from competing with classmates that cheat in general? You also had parents that do most of the tasks for their kids since forever
What makes LLMs stands out of the pack of other forms of assistance (parents, older siblings, software, online solutions, etc.)
This isn’t a competing with kids issue, though. A kid handicapped by ChatGPT will fail in-class examinations.
To me it sounds like you are worried more about the competition for resources than giving your daughter an education.
she wont need a chatGPT, she will develop one in her nervous system, like a normally functioning human
That's not really true. At least in my experience.
Since I graduated nobody cared about my grades at all. On the other hand everyone cared if I can do stuff.
When I was applying for Uni my grades mattered very little and my performance on standarised and proctored tests mattered a lot.
I guess it might be different where you have grown up?
That's called "winning the battle and losing the war".
I haven't understood the GGP's point of view until your comment. Those kids are competing by grade, not by knowledge!
This is messed up to a different level than a lot of people here will expect.
And corporation can be even simpler if you get in right types of roles, where no hard skills might be needed.
An idiot may be able to fake being a medical doctor with the right AI, but only until someone realizes they can fire the useless human.
Parents income and zip codes impact educational attainment more than class rank outside of the oligarchy anyways.
Opportunity and social mobility is far more restricted in the US than most people realize.
This is in part why the college debt is problematic, valuing a piece of paper over the education itself is already a bubble that has popped.
1) Do you have a comparison with European countries that makes you think the US is worse?
2) What do you think US mobility numbers should look like? Pure randomness is not desirable either.
I clearly don't understand how difficult life is outside the US. Yes, it is important to know the material. However, it is also important to know your surroundings and what everyone else is doing. If you are in India and literally everyone else is cheating in exams, well you better start doing that as well just to keep up. If you are gen alpha and everyone else is using ChatGPT to polish essays, ...
And that's not to say the behavior doesn't exist in the US, just maybe in a different social strata or subculture than you're in. Specifically, the phrase "tiger mom" entered our lexicon due to a book written by a (ethnically) Chinese woman, but is not limited to it.
What we need to start focusing on is teaching children how to use AI to evaluate their solutions. We also need an AI built for students that can mentor children and help them develop their own solution.
Shielding children from the technologies that's going to be part of their world doesn't seem wise. They'll push the technology in new and unexpected directions.
A lot of the system is based off of memorization and paperwork anyways. Something an AI can do all day, every day.
I see such competition as hare vs tortoise races, where the students who get an early unfair advantage end up cheating themselves out of an education, in the sense that all they come to know is how to prompt an LLM, rather than break down and solve hard problems.
It's still heartbreaking to consider both the early discrimination she will likely encounter, and the later failures that will anticipate her classmates, and the society that they will come to inhabit as adults.
i came here to laugh at this thread.. no different than calculators and google both of which were misused in the same way. and then people who are "worried" about it and use that as an excuse to do it themselves. and now you know why the tech industry sucks
Grading on a curve is very much a thing. Rarer in lower levels but not completely unheard of.
Even if it weren't GPA is still used for ranking students for opportunities like college or summer programs with limited slots. The effect of GPT'd homework depends on how well they perform on the tests though but some classes are predominantly take home work or at least were when I was going through school.