Comment by GPerson
2 days ago
The kids who used AI and studied for a similar amount of time as the non-AI high performers apparently had similar (slightly higher) performance. The kids who used AI and did poorly used the AI to do the homework for them (this is what the article says).
I believe AI is basically an amplifier of bad and good. I’m cynical about the world and assume it will be used more for bad than good, but I don’t doubt some of the best people in every field will be using AI to amplify their work in good ways.
It's not so clear that AI is an amplifier.. The paper has some fascinating analysis on this topic:
"At the other end of the distribution, AI students who spend more than 65 minutes on their homework receive homework and exam scores similar to those of non-AI students, suggesting that these students do not use generative AI for homework assignments. However, this group consists entirely of students who adopted generative AI no more than Öve months. Six months after adoption, no AI student spends more than 65 minutes completing their homework (see Figure A5). This is consistent with the gradual process of learning how to use AI tools. It also suggests that AI crowds out the highest level of e§ort."
"Interestingly, in the range of 50-65 minutes, the median and the interquartile range of exam scores of AI and non-AI students are similar. This implies that, in the range where AI students and non-AI students have overlapping homework times, students who spend the same amount of time completing homework on average receive similar exam scores."
"This pattern shows that students who spend the same amount of time on homework learn similarly, with or without generative AI. In other words, generative AI reduces time spent learning for the majority of AI students but not learning efficiency for those who spend the same time studying as the non-AI students."
Depends who's using it. Like many tools the force multiplier depends on the operator.
I'm confident it's an amplifier for people who know how learning works and already do a lot of it, successfully. However the level of "learning fluency" I'm talking about isn't reached for many until late college or grad school, and sometimes not at all. So I'm not surprised by the quoted results for 12-18 year olds.
And people who want to learn. Most teenagers lack agency in their studies. They aren't in high school because they love it but because they have no choice.
Sometimes things are just common sense pure and simple.
1 reply →
It would be good to see the effect of access to AI during preparation on those who previously achieved top 10% points in exams of similar topics before. I suspect they would benefit further.
My apologies for coming off as over-enthusiastic, I am currently obsessed with this study. Here is another quote:
"The negative learning effects are larger for students with higher initial achievement. The differences in the estimated full (6-10 month average) effects are substantial, with a 50% gap between the most negative effect (-24 percent) for the highest tercile and the least negative (-16 percent) for the lowest tercile. "
Not top 10% as you asked, but the closest to what you asked. My working hypothesis is that top performance is highly correlated with willingness to work hard, and AI decreases the motivation to work hard.
1 reply →
well most university exams are designed to measure how much you study. so we didn't really need a study to tell us, "Exams continue to measure what they are designed to measure."
they're not designed to measure general aptitude, or function as admissions criteria, or screen for job applications, or any other numerous things they are used for.
there can be many questions of pedagogy. one of them is, what do our exams measure and how do we use them? professors who say, "My exam is designed to measure who studies, not be used for all these other purposes that they are actually used for" - I don't buy it. It's the same as late night comedians saying they are not responsible for solutions, even when spending 90% of their air time making political jokes.
THIS is the pedagogical issue, that pedagogy has NEVER caught up with the scope of responsibilities. This is acute in STEM - I mean, the humanities departments are generally pretty well run, all things considered, in this regard. Generative AI is accelerating that pre-existing crisis.
The fact that exam scores are correlated with how much you study is not the same as exams only reflect how much you study. Two students who study the same amount could have very different exam scores. The reason that there still is a strong correlation between exam score and time of study is because if all other things being equal, students who studies more have higher exam scores.
2 replies →
> well most university exams are designed to measure how much you study.
Huh? They're designed to measure how much you know. They can't see how much you study, nor would they have reason to be interested.
4 replies →
The scary thing is that 81% of AI users in this study were determined to be "outsourcing" their homework to the LLMs - and the rate increases the more exposure they had to LLMs.
The "slightly higher" performance is based on statistically insignificant samples (between 4 and 20 students, depending on the context, out of the total population of 26,000): https://bsky.app/profile/benjaminjriley.bsky.social/post/3mt...
I thought that it was generally accepted by now that homework in the volumes that it is being assigned in the modern day was not found to be beneficial in any significant way in the first place? Maybe once all students start outsourcing it to AI it might finally die like it deserves to. People these days grow up with almost no free time for themselves, it's all school, sports/extracurriculars or homework nonstop. All worker drone and no play makes for an increasingly dysfunctional society.
What I remember hearing is that exercises and practice are needed to transform words on the page to something you know and master.
What I remember being questioned is does it make sense to do those exercises as homework or would they be better in school.
Or on the flip side should school get out early like 11 or noon, like I think the german gymnasium does and have all the exercises as homework.
The US system where children get out of school at 15:30 and still have a bunch of homework seems a little lopsided someplace.
The 3 month break in-between school years is definitely questionable.
I think the answer is a complex one because it intersects with personality and neurodivergence.
Depending on who you are and your family situation any form of homework can be a real challenge. Not because of what you are studying but because of how difficult it is to sit down and do anything you aren't passionate about. Certainly anyone with an executive function disability will have that challenge.
> I thought that it was generally accepted by now that homework in the volumes that it is being assigned in the modern day was not found to be beneficial in any significant way in the first place?
Citation needed? I have no clue where you got this from. I hadn't even heard of it as a conjecture, let alone as something anyone accepted, let alone as generally accepted...
2 replies →
How do you expect students to learn anything when:
1. They don't do any homework.
2. All the in-class time is split between the teacher babysitting and playing social worker to problem students, and lecturing, with little to no opportunity to actually practice what they've learned?
I understand that some students don't have home environments that are conductive to doing homework well. I understand that some students are enrolled in five hours a day of extracurricular university-application-padding activities. I understand that some students have incredibly poor screen discipline and impulse control.
But I don't understand that anyone has magically figured out how to teach complicated things to students, and have it stick without them spending a lot of time practicing what they are learning.
As anyone who has tried to do something hard knows, the first step to being good at something is to spend a lot of time being pretty shit at it.
A student who has written and received feedback on 500,000 written words is going to be way better at writing than that same student who wrote 50,000, just like someone who has put 5,000 hours of focused practice into playing the piano is going to be better than my dumb ass, who has only put 100 hours in.
(If you found the solution to get good at stuff without practicing it, I'd love to get good at piano without putting any homework in on it.)
All play and no work makes for an even more dysfunctional society. Kids need to do their homework, both to train their minds but also to develop integrity and work ethic. Being able to diligently work toward a distant goal is not something you're born with.
This is kind of damning, isn't it? "Similar or slightly higher" for spending the _same time_ means it's not, in fact, helping. At the same time as being extremely damaging to a significant number of students who, of course, use the AI to cheat. Because everything about it makes cheating easy.
> I believe AI is basically an amplifier of bad and good.
Same can be said of technology in general tbh.
You aren’t being cynical, you’re arguing with disingenuous entities with money on the line. We already know it’s used primarily for the negative case. Everyone who ever intended High School or College knows this.
Definitely had a sheltered public school experience, took AP stats my senior year and realized all the top students were sharing answers from an earlier period through text messages. After figuring this out I too joined in on the action, then shortly after I quickly deskilled.
Suppose it's good to learn how elastic the brain is, in both directions, at a young age where it doesn't matter.
> at a young age where it doesn't matter.
Doesn't it matter the most at a young age?
The dangerous thing about amplifying is that bad people are often more willing to amplify their activities, because they don't care about the negative effects. We are seeing this now as AI companies (and large companies of all kinds) rush to secure whatever advantages they can, regardless of the negative externalities. Meanwhile people who actually care about doing the right thing get trampled.
We need to shift the incentives by adding ruinous penalties for things that are currently quite commonplace if they are done by large players. Some dude training his own AI on his own computer can scrape and train. The fine for OpenAI or Meta using a single copyrighted book without permission should be in the tens or hundreds of millions.
What we're seeing currently in our society is a "loophole inversion" where the rules have an effect mainly via their loopholes. The most profitable activity is to find loopholes and exploit them as frenetically as possible to gain as much advantage as you can before the loophole is closed, or get people hooked on the loophole so it's retroactively legalized. Entities that are big enough to do this are big enough because they have lots of money behind them. Entities doing the same kinds of things without lots of money are not really doing much harm. So the best approach is to adopt a "sliding scale" in which even tiny violations by wealthy actors result in penalties enormously greater than fairly large violations by small players.
It's a familiar story. If you copy/paste Wikipedia and turn it in, you learn nothing. If you skim Wikipedia, hit up the references, the consult other sources, then synthesize your own thought, you probably get farther faster than you would without it.
We could probably cook up thousands of examples of the same problem (YouTube DIY tutorials, GPS navigation, etc.)
This is my thoughts as well! It makes smarter people smarter and dumb people dumber!
There is zero evidence for this.
LLMs make significant mistakes frequently and smart people have no way of judging those mistakes outside their domain expertise. They are also sycophantic and great at being an echo chamber which makes people feel smart even if they are not.
So I think the burden of proof is on you to prove that they somehow amplify intelligence, it seems highly unlikely.
Most domains have some kind of internal consistency/theory building you can do. A smart person can certainly notice inconsistencies when trying to learn something. In fact they're likely to be points of confusion that the smart person will dive into just to try to make sense of things, even if they don't suspect the LLM is at fault.
No evidence but somewhat of a counterpoints:
Smart people know LLMs confabulate and tell them they’re Absolutely Right! Smart people don’t want to be embarrassed by trusting the hallucination machine and revealing their gullibility to others.
Why would a smart person go to an LLM for an answer they cannot judge or test, be succeptible to flattery and sycophancy rather than picking up on the emotional manipulation and being suspicious/sceptical of the interaction, or looking for support from an echo chamber rather than a Socratic opponent?
All of those sound like flaws and defects of dumb people?
1 reply →
All at the cost of ... {List of negatives regarding the construction and powering of AI data centers }
The social costs are worse.
Most importantly, it makes investors think dumb people are smart.
It helps smart people be barely more effective and helps dumb (more importantly, people who do not value effort, people who are lazy, people who are self absorbed) people shit out endless streams of worthless tokens that can swamp out everything.
Raising the noise floor like this only makes it that much harder to find "Smart" people, which we were already doing terrible at.
I use Claude every single day, but this is such a bad tradeoff. Maybe it will help me standup a quick fix when that is needed. Maybe it can help me dig through documentation to find relevant bits and figure out the unstated assumptions underlying it. Maybe it helps me generate test cases.
Meanwhile, my day to day life is now noise. All social media is noise. All content is noise. Slop pours onto me from all directions. Writing more test cases isn't helping me.
Am I smart? Am I dumb? I don't care, right now I'm deafened
All I've noticed is AI creating a perverse incentive to make everything as complicated and bureaucratic as possible, so only the people that know how to leverage AI to cut through it ever succeed.
I'm using Claude at work myself and am impressed with the product, but notice that this is the only reason I need to use it at all. Our product pages were shit to begin with, now they're AI-generated and somehow even worse. Our procedures are incomprehensible spaghetti with enough arbitrary context switching to give a sadistic Soviet municipal administrator an erection at the thought of watching anyone try to actually follow them.
Use AI to create inefficiencies, then use AI to bypass them. Those who can't do the latter will struggle to survive.
> Raising the noise floor like this only makes it that much harder to find "Smart" people, which we were already doing terrible at.
Clarification: to value “smart” people, which we were already doing terrible at.
> Raising the noise floor like this only makes it that much harder to find "Smart" people,
It does give us a new heuristic, though: people who are willing to completely cut generative AI out of their lives (cold-turkey, if you ever started using it) are a much smaller group of, predominantly thoughtful, people. You do have to give up Claude to be part of this group, but from what you say, that's no great loss, and no longer being deafened is worth it.
This has considerable advantages over conventional elitism, because the barrier-to-entry is negative in almost all cases.
The one exception I've found is assistive tech, where the state-of-the-art is so poor that vibecoded slop is genuinely an improvement over the state-of-the-art, and in many cases the tooling simply isn't available to make your own assistive tech (unless you want to bootstrap an entire networked computing environment, which isn't very helpful when you want to do your online banking and do not, in fact, work at your bank).
But there are not many principled exceptions where you could seriously argue that the trade-off is worth it. Take mathematics, for example, which we often see touted as a "good use-case" of generative AI. The primary advantage of generative AI in mathematics is being able to search though a vast corpus of ivory towers and inconsistent terminology (without proper attribution) to locate and connect ideas that can help solve problems. The deficiency this is addressing is elitism, inadequate communication, and inadequate indexing within academic mathematics. This problem is entirely created by the academic mathematicians, and has been known for nearly a century (per https://news.ycombinator.com/item?id=49392265 points out one such.)
Vibe-coding assistive tech which doesn't yet exist, as a temporary scaffold to improve the quality-of-life of yourself and others in a social world dominated by non-essential access barriers is, to my knowledge, the only exception to this principle that can be justified. If you treat people who make other excuses, or who don't even bother with excuses, as not worth listening to, you lose little – and doubly-so, if you make your stance clear, so that others know the "cost" of gaining your attention.