Search and filter have been the paradigm for a while, since the book index was invented.
What are the tools of thought that will help students now succeed and advance technology in 20 years?
Perhaps there is a market fit for a laggard GPT. Something useful but exhibits clear errors that a student could pick up on, and so recognize the risk of blindly accepting information like this. The same issues are associated with excess reliance on traditional publications, "the armchair scientist" and so on.
Maybe, trained sequentially on the curriculum like a student, or something personally orthogonal so that the bot excels where the student is known to fail, create gaps where the student is known to succeed. It seems straightforward to fine-tune in this way since the corpus is smaller. For each grade level and setting A/B/C across the 4 subjects, maybe some 300 fine-tuning?
100% of the students I have been teaching (roughly 5 to 15 years old) are using chatgpt for almost everything. It's sometimes a struggle to get them to think for themselves!
This is the sequel to "You better learn calculus because you won't always have a calculator in your pocket." Well I already had a calculator on my Casio watch and now I have access to an AI supercomputer my pocket. Check mate Ms. Levenstein, joke's on you.
>It's sometimes a struggle to get them to think for themselves!
Was it any easier/better before ChatGPT? In my generation MTV and DOOM was to blame for kids being unable to think for themselves. Before that, rock metal was to blame for perverting the youth.
An LLM is not an AI supercomputer. It will regularly give you information that is false, and if you do not know the subject well, you probably won't notice how wrong it is. We are still a very long way from Star Trek's computer being able to help you out.
Hard to dive, if you don't know how to swim. LLM generated information needs you to be able to understand what it's producing, to be able to judge it. For that, you need education... That it can't trustworthily provide.
You can't expect someone without a background to understand quantum entanglement. Can't expect someone without the knowledge to comprehend memory management.
And if you do have the background... You're going to do a much better job than AI "slop", and that nickname has become popular for a very good reason.
A calculator won't give you bad information if you use it correctly, though. AI, will. The same prompt can generate vastly different answers.
Even the codegen examples given by the AI companies themselves have flaws in them. Critical flaws, like Claude's testing rig that doesn't test what it says it does, for example. The system is inherently flawed for most purposes it is currently being used for.
A calculator won't give you bad information if you use it correctly, though. AI, will. The same prompt can generate vastly different answers.
I think this is exactly it except that part of “knowing how to use it” is also largely about knowing its limitations… ‘trust’ but verify my 11-year old has been using chatgpt/claude since it came out and I have nothing but awesome experience with how she is using it
You sorta just made the point I'm trying to make here.
How the heck, does your 11-year old, verify? Do they turn to you, who already has the necessary background? The AI generated information cannot be verified by someone who cannot already do, what it does.
how do you verify? think about how would you do that if say you were forced to use AI for everything you do (I have friend that works in a place like this…).
of course my kid is 11 so she is learning about algebra and electro motors and logic and roman empire… “AI” is one step in that learning but an insanely patient teacher. back in a day - she did not understand cube roots… she asked, was given an answer, she still didn’t get it, was given an alternate explanation which still didn’t click (worse than original), asked again (“I still do not understand, do you have another way to try to explain it to me…”) and so on. again, it is a guide, a very knowledgeable and patient guide… it is part of the journey, not the destination. anyone using it as destination (by reading MANY HN posts this is a vast majority of people) is going to be in the world of hurt
I think technology is making the general populace stupider. Presidential debates are a litmus test for what politicians expect of the populace. A hundred years ago they sounded a lot different than how they sound now. Does your nihilism encompass this change as well?
My tone might have indicated otherwise, but I don't have a solution because I don't really see a problem along this axis. I don't really care about how stupid everyone is, as long as their material conditions are at a standard I consider good.
> This is the sequel to "You better learn calculus because you won't always have a calculator in your pocket."
Yeah it's all fun and games until you find a cashier who can't substract 20ct from $1 and need to get their iphone out to solve this really hard equation. Or people who can't locate Africa on a world map. You don't build a prosperous society on these foundations, regardless if LLMs can answer your questions
>Yeah it's all fun and games until you find a cashier who can't substract 20ct from $1 and need to get their iphone out to solve this really hard equation.
WTH, Cash registers have already been doing that automatically at every transition for the past 70 years or so.
>Or people who can't locate Africa on a world map.
How does that knowledge have anything to do with their work lives? Does your boss jump in the middle of your TPS reports
in Jira and ask you to point Africa on the map within 3 seconds without using your phone or you're fired?
I lived in the 90's Eastern Europe where a lot of people knew where Africa was on the map but society was anything but prosperous. Knowledge of basic geography is no guarantee of prosperity, your ability to get money is, and plenty of people in rich countries can make money without knowing where Africa is since workers in rich countries are way more specialized than before. Now you can be a car owner without knowing how to change the oil or how a clutch works. That was not the case 50 or so years ago.
Knowledge of random shit or trivia has no value anymore in developed societies where every worker becomes more and more specialized in specific niches versus societies where everyone knows a little bit of everything from car repair, a little bit of plumbing, some farming and maybe some medical patch work on the side but nobody is specialized in any valuable skills.
Critical thinking makes one enlightened, not memorizing trivia that anyone can Google on a whim, like where Africa is on the map.
Understanding how democracy, taxes, politics, resources, compounding interest, inflation, lobbying, corruption, etc and ability to connect the dots between these, is more important for society than pointing at foreign continents on the map.
> Understanding how democracy, taxes, politics, resources, compounding interest, inflation, lobbying, corruption, etc and ability to connect the dots between these...
So how do you do this if you have an oracle that answers the exact question you have without any struggle? I find that the most useful things I learnt in life I bumped into after taking wrong corners while looking for something else.
>So how do you do this if you have an oracle that answers the exact question you have without any struggle?
By that logic we should all go back to using an abacus and slide ruler for calculations.
Life is about making progress, not struggling with outdated ways for the sake of it. Struggle isn't a virtue.
If give people the task of banging nails into a wall and one uses his head to bang nails, the other use a hammer, and the third uses a nail gun, which is more productive and valuable, the one who struggles the most or the one who gets it done quickly and efficiently?
> How does that knowledge have anything to do with their work lives? Does your boss jump in the middle of your TPS reports in Jira and ask you to point Africa on the map within 3 seconds without using your phone or you're fired?
> Knowledge of random shit or trivia has no value anymore in developed societies where every worker
Yeah I mean if your entire existence revolves around your job and what your boss thinks of you we're in for a weird discussion... Life has more to offer than 50 year of work and 15 years of retirement. What a strange and sad way to envision life. Cultivating your mind and knowledge, as well as your body, should be way up in the list of things we have to focus on, you might want to check the etymology (if you know what that means, it's not a very useful term in the workplace) of "education"
> Now you can be a car owner without knowing how to change the oil or how a clutch works. That was not the case 50 or so years ago.
Nice, that's definitely what we want, monkeys pushing buttons without any ideas of what's going on under the hood. Why would you want to know why the earth is round, what gravity is, why braking distances are longer in rain, how ABS works, what is engine braking, why you can't stomp on the brakes for 30min straight while driving down a mountain road, ... People smoke their clutches all the time, run on bald tires, use summer tires in winter, pour oil in their coolant tank.
What's the point of being ultra specialised in a niche field, or even worse, profession, if you're a complete dummy in life
edit: and also, most people aren't even specialised, so now you get the worst of both world
What an entitled and 'holier than though' view. When you can't afford rent, groceries, education or medical bills due to inflation, and don't have a support network or family, what use is it to the Average Joe to be "enlightened" on random shit like pointing Africa on the map.
Maslow's hierarchy of needs (shelter, food, sex, relationships) will always take priority for most individuals over learning random pub-quiz trivia.
Knowing where Africa is located is "enlightenment"... this forum keeps on giving
All the problems you listed come from the same root, and LLMs won't solve them
> Maslow's hierarchy of needs (shelter, food, sex, relationships) will always take priority for most individuals over learning random pub-quiz trivia.
Time to plug in your brain and ponder why the top economies of the world aren't able to provide basic human needs while generating more wealth than ever before, once again, same root as above, once again, LLMs won't solve any of that.
We could have AGI and immortality tomorrow, as long as we keep the free for all ultra individualistic profit seeking capitalistic system it's not going to change jack shit.
> random pub-quiz trivia.
> Maslow's hierarchy of needs
Wow slow down cowboy, "maslow's hierarchy" IS random pub-quiz trivia, get that off your brain real quick, you wouldn't want to become enlightened do you ?
>Comfort levels with using ChatGPT for different types of assignments vary among students: 54% found that using it to research new topics, for example, was an acceptable use of the tool. But only 18% said the same for using it to write an essay.
Not sure how old you are, but WolframAlpha has been exactly that for a solid decade and a half. (It was a little late for while I learned calculus, but I used it for reference afterwards.)
Also, I’m curious - are GPTs/LLMs actually able to provide consistently accurate explanations and solutions for calculus?
They’re historically notoriously bad at even basic arithmetic and algebra, and a quick search seems to confirm them performing similarly with calculus, but I’m open to being wrong and things having improved on that front if you have some evidence.
Finally, echoing the other commenter, you really shouldn’t need a GPT-type tool to do math. If you do, you arguably don’t actually understand the concepts. And I’d honestly be shocked if a LLM could break things down well enough to genuinely help someone who doesn’t understand how to do a calculus problem, having myself tutored friends who didn’t understand things back when I was in school (and using LLMs extensively for my own work).
Some of the best math classes I took didn’t grade homework at all, only exams for which we only had pencil and paper (or some weighted homework as 10% or less of the final grade). But of course, if you didn’t do all the homework, you’d never pass the exams.
I feel too many people treat school as just a place you learn stuff, where in reality the most important part imho is the challenge of learning, that's what sticks.
I have a degree in chemistry, I almost forgot everything about chemistry, but I have very huge advantages when it comes to learning complex topics compared to those that have the explanation handed to them.
In Italy you could not fail a school year if you failed only 3 topics, so the little engineer in me figured out I could simply ignore the existence of mathematics and physics for 5 years and just do the rest.
In university I had to go through calculus 1/2, complex analysis, quantum mechanics, etc, etc.
So as soon as I got into calculus 1 the first semester of the first year I bought several books on the topics and just went page by page understanding and thinking about the definitions and did 100x the exercises of my class mates. By the end of it maths have never been a problem in my life.
Whatever I need to learn I am now equipped with the tools to do so starting from the basics and this pays off in many aspects of my life.
Seriously. The example in the book goes A->C->F to get the answer and I needed the TA to explain how you get to C from A and then also the steps to get to F from C in order to understand better then merely filling in the numbers.
Well, I dunno. I have a 17 year old in his last year of high school. After the initial novelty wore off, I don't think ChatGpt is used much anymore, not by him at least.
- You can hardly use the writing in an essay, as it is replete with listicles. You'd have to completely rewrite the stuff anyway. Moreover, he know he'll be tested on this stuff, and yeah that does seem to have an impact.
- Even using ChatGpt for explaining stuff is hardly used. For sure, I use it much much more often than he does. Also here it seems that it is hardly worth it, as the ChatGpt answer will differ too much from what is expected on the test. He much prefers to study with last years' tests--and the accepted answers-- than to use ChatGpt.
While it's not necessarily great for writing an essay from scratch, it is a pretty good editor of an essay. My daughter uses it to 'grade' all her essays before handing them in (With a prompt like "This is a 2 page essay about X for a 10th grade social studies class, what grade would it get") and then follows up with "what needs to be improved to get the essay from a B to an A". And based on what I've seen it does a pretty good job and gives quite reasonable recommendation that are appropriate for the grade and class in a question.
Using it to explain stuff. Don't we already have books written by professionals with hopefully years of experience? And these are given out to the students... And anything in them is taken as truth for the tests...
In context of school, I would think just reading the book is lot simpler than figuring right thing to ask.
Ofc, there is situations like literature where asking the ai to tell about book is probably easy cheat... But you also would have wikipedia and whatever...
Don't we already have books written by professionals with hopefully years of experience? And these are given out to the students...
Do you remember your school textbooks? At best you could say they were optimised for some sort of median student that prefers the 'standard' way of acquiring information. If that's not you, or you want to learn information outside the textbook you are out of luck. LLMs have the potential to be the infinitely patient teacher that will explain the things over and over in as many different ways as necessary for each student to fully understand the information.
Beyond that, from about 6th grade or so we were definitely expected to start going beyond our textbooks for some assignments. Going to the school library or taking my bike to the local library to work on assignments was not uncommon.
By the way, of students who have heard of ChatGPT, 32 percent have used it.
But more importantly, this feels like a broad thing. There's a huge difference between 'hey I have to write a paper on the Roman economy, what are some sources about that' and 'write a paper about the Roman economy'. Similarly a difference between 'crap, what's the formula for the volume of a sphere' and 'calculate the volume of a sphere with diameter 12 feet'. Personally I would rank the 'cheatiness', from less to more, as sources < formula < writing a paper/calculating.
This whole study feels like it hinges on what students consider 'help', since it was self-reported. If you try to get it to help but it only turned up sources you knew, does that count as helping? If you tried but it spat out an awful paper and you got a really bad score, does that count as helping? If you wound up having to spend more time correcting its output than you would've just writing the paper, does that count as helping? It's just so variable depending on individual standards the study feels kind of worthless as an actual indicator of stuff.
One kid of mine is 12 in Grade 7 here in Ottawa, Canada and all of her friends apparently use ChatGPT on occasion. They are not using it for everything, but they are using it at least periodically.
Are you in the "it's just like with calculators" camp? I think there's a fundamental difference between a tool that just simplifies some very specific solution steps vs a tool that takes the literal assignment question, regardless of how "updated" or "more difficult" it is, and outputs an answer that you can probably submit without even checking for a passing grade. Watermarking, guardrailing etc. won't fundamentally fix this, there will always be access to more unfiltered models.
> Cheaters, by definition, avoid real work, so they won't be using custom models unless the stakes are sufficiently high.
I think it's a non-sequitur, some might get more enjoyment out of tinkering with LLMs than out of whatever the assignment is asking you to do. But even if you don't believe that, it should be obvious that the ones using the tool needn't be the same as the ones putting in the work.
This isn't yet another "we need to change a few things at the tail end" problem like internet was. You have people getting in higher education with barely any skills whatsoever, not even the basics to just get by
It's going to be a major issue, especially in a world that requires higher and higher education, coupled with systems which are more and more complex, and fragile. Internet was supposed to make everyone smarter, we quickly discovered that information != intelligence, let's see how it goes for LLM but if anything it'll make use even more dependent on technology.
And couple that with teens spending on average 5 hours per day on social media...
I sometimes watch archive footage of kids answering questions in school in the 70s vs now, and it's painfully obvious that the level plummeted, the average kid back then was more eloquent than most young adults today
> It's going to be a major issue, especially in a world that requires higher and higher education
Society will bifurcate further; as mental labor is offloaded to GPUs, the demand for education will drop. A greater percentage of the workforce will be doing manual labor, and that doesn't really require much of an education at all. Those with means will have access to a proper education as always, but that group of people is going to shrink.
These models are far better teachers for my kids than their actual teachers. they are far more engaged, able to go down rabbit holes guided by their own curiosity, and end up learning more.
I do my best to ensure they're not using these models to do their work for them, but they are invaluable teaching tools, nonetheless.
Hopefully this will be one more nail in the coffin for the sprawling public education complex.
Why "learn" anything when the computer just gives you the answer?
The computer won't chide you for not manually gathering and reviewing sources for an essay, not understanding units in physics formulas to check your own answers, not extracting casual relationships between events by closely reading a dense text. It'll give you the answer as quickly as possible so that you can go back to viewing short videos.
All the information has always been out there for students to pick it up but a vanishingly small few actually do.
What are the tools of thought that will help students now succeed and advance technology in 20 years?
Perhaps there is a market fit for a laggard GPT. Something useful but exhibits clear errors that a student could pick up on, and so recognize the risk of blindly accepting information like this. The same issues are associated with excess reliance on traditional publications, "the armchair scientist" and so on.
Maybe, trained sequentially on the curriculum like a student, or something personally orthogonal so that the bot excels where the student is known to fail, create gaps where the student is known to succeed. It seems straightforward to fine-tune in this way since the corpus is smaller. For each grade level and setting A/B/C across the 4 subjects, maybe some 300 fine-tuning?
reply