Putting aside the ludicrous confidence score, the student's question was: how could his sister convince the teacher she had actually written the essay herself? My only suggestion was for her to ask the teacher to sit down with her and have a 30-60 minute oral discussion on the essay so she could demonstrate she in fact knew the material. It's a dilemma that an increasing number of honest students will face, unfortunately.

I think that AI has the possibility of weakening some aspects of education but I agree with Karpathy here. In class work, in person defenses of work, verbal tests. These were corner stones of education for thousands of years and have been cut out over the last 50 years or so outside of a few niche cases (Thesis defense) and it might be a good thing that these come back.

It ended up being harder then writing an ordinary paper but taught us all a ton about citation and originality. It was a really cool exercise.

I imagine something similar could be done to teach students to use AI as a research tool rather then as a plagiarization machine.

I've seen assignments that were clearly graded by ChatGPT. The signs are obvious: suggestions that are unrelated to the topic or corrections for points the student actually included. But of course, you can't 100% prove it. It's creating a strange feedback loop: students use an LLM to write the essay, and teachers use an LLM to grade it. It ends up being just one LLM talking to another, with no human intelligence in the middle.

However, we can't just blame the teachers. This requires a systemic rethink, not just personal responsibility. Evaluating students based on this new technology requires time, probably much more time than teachers currently have. If we want teachers to move away from shortcuts and adapt to a new paradigm of grading, that effort needs to be compensated. Otherwise, teachers will inevitably use the same tools as the students to cope with the workload.

Education seemed slow to adapt to the internet and mobile phones, usually treating them as threats rather than tools. Given the current incentive structure and the lack of understanding of how LLMs work, I'm not optimistic this will be solved anytime soon.

I guess the advantage will be for those that know how to use LLMs to learn on their own instead of just as a shortcut. And teachers who can deliver real value beyond what an LLM can provide will (or should) be highly valued.

The key issue with schools is that they crush your soul and turn you into a low-agency consumer of information within a strict hierarchy of mind-numbing rules, rather than helping you develop your curiosity hunter muscles to go out and explore. In an ideal world, we would have curated gardens of knowledge and information which the kids are encouraged to go out and explore. If they find some weird topic outside the garden that's of interest to them, figure out a way to integrate it.

I don't particularly blame the teachers for the failings of school though, since most of them have their hands tied by strict requirements from faceless bureaucrats.

It's getting rid of cheap methods.

Scantrons and bluebooks were always a way that made it cheap for institutions to produce results. Now those methods kinda seem silly, right?

500 person freshman lectures seem kinda absurd now, right?

Teaching via adjuncts that had 3 days notice for the class and are paid nothing is kinda scammy, right?

R1s professors whose tenure evals have nothing to do with teaching is kinda wrong, right?

The Oxbridge model of 5-10 person classes with a proctor is what the education with AI is going to be about. It's small, intimate, and expensive.

The problem is that the structure pushes for teaching productivity which basically directly opposes good pedagogy at this point in the optimization.

Some specifics:

1. Multiple choice sucks. It's obvious that written response better evaluates students and oral is even better. But multiple choice is graded instantly by a computer. Written response needs TAs. Oral is such a time sink and needs so many TAs and lots of space if you want to run them in parallel.

1.5 Similarly having students do things on computers is nice because you don't have to print things and even errors in the question can be fixed live and you can ask students to refresh the page. But if the chatbots let them cheat too easily on computers doing hand written assesments sucks cause you have to go arrange for printing and scanning.

2. Designing labs is a clear LLM tradeoff. Autograded labs with testbenches and fill in the middle style completetions or API completetions are incredibly easy to grade. You just pull the commit before some specific deadline and run some scripts.

You can do 200 students in the background when doing other work its so easy. But the problem is that LLMS are so good at fill in the middle and making testbenches pass.

I've actually tried some more open ended labs before and its actually very impressive how creative students are. They are obviously not LLMs there is this diversity in thought and simplicity of code that you do not get with ChatGPT.

But it is ridiculously time consuming to pull people's code and try to run open ended testbenches that they have created.

3. Having students do class presentations is great for evaluating them. But you can only do like 6 or 7 presentations in a 1 hr block. You will need to spend like a week even in a relatively small class.

4. What I will say LLMs are fun for are having students do open ended projects faster with faster iterations. You can scope creep them if you expect expect to use AI coding.

A much bigger question is what to teach assuming we get models much more powerful than those we have today. I'm still confident there's an irreducible hard core in most subjects that's well worth knowing/training, but it might take some soul searching.

This topic has been an interesting part of the discourse in a group of friends the past few weeks because one of us is a teacher who has to deal with this on an almost daily basis and is struggling to get her students to not cheat and the options available to her are limited (yes, physical monitoring would probably work but requires concessions from the school management etc. it's not something that has an easy or quick fix available.)

[0] https://oxide-and-friends.transistor.fm/episodes/ai-in-highe...

That is just such a wildly cynical point of view, and it is incredibly depressing. There is a whole huge cohort of kids out there who genuinely want to learn and want to do the work, and feel like using AI is cheating. These are the kids who, ironically, AI will help the most, because they're the ones who will understand the fundamentals being taught in K-12.

I would hope that any "solution" to the growing use of AI-as-a-crutch can take this cohort of kids into consideration, so their development isn't held back just to stop the less-ethical student from, well, being less ethical.

School is packed with inefficiency and busywork that is completely divorced from the way people learn on their own. In fact, it's pretty safe to say you could learn something about 10x by typing it into an AI chat bot and having it tailor the experience to you.

Schools need to become tech free zones. Education needs to reorient around more frequent standardized tests. Any "tech" involved needs to be exclusively applied towards solving the supply and demand issue - the number of "quality teachers" to "students per classroom."

I admire Karpathy for advocating common sense, but none of this will happen because SV is full of IQ realists who only see "education" as a business opportunity and the bureaucratic process is too dysfunctional for common sense decisions to prevail. The future is chrome books with GPT browsers for every student.

If my son should grow up to run into the same kinds of cognitive limitations, I really don't know what I will tell him and do about it. I just wish there was a university in a Faraday cage somewhere where I could send him, so that he can have the same opportunities I had.

Fun fact on the side: Cambridge (UK) getting a railway station was a hugely controversial event at the time. The corrupting influence of London being only a short journey away was a major put-off.

Zero homework grades will be ideal. Looking forward to this.

She started grading conversation than the students have with LLMs.

From the question that the students ask, it is obvious who knows the material and who is struggling.

We do have a custom setup, so that she creates an homework. There is a custom prompt to avoid the LLM answering the homework question. But thats pretty much it.

The results seems promising, with students spending 30m or so going back and forth with the LLMs.

If any educator wants to Ty or is interested in more information, let me know and we can see how we collaborate.

Take away the internet. Except in a research/library scenario. Give them a limited time to complete tasks. This would promote a stronger work ethic, memory/recall and more realistic to time management skills. They need to learn to rely on themselves, not technology. The only effective way is to remove tech from the equation, otherwise the temptation to cheat to compete/complete is too strong.

I'm not minimizing Karpathy in any way, but this is obviously the right way to do this.

Calculator analogy is extremely inaccurate, understandably people keep doing this comparison. The premise is that calculator didn't take bookkeepers' job, but instead it helped them.

First of all calculator do one job and does it very well, you never question it because it solely works with numbers. But AI wants to be everything, calculator, translator, knowledge base etc.. And, it's very confident at everything all the time until you start to question it, and even then it continues to lie. Because sadly current AI products' purpose isn't to give you accurate answer, it's about making you believe that it's giving you credible information.

More importantly calculators are not connected to the internet, and they are not capable of creating profile of an individual.

It's sad to see big players push this agenda to make people believe that they don't need to think anymore, AI will do everything for them.

Learning how to prepare for in-class tests and writing exercises is a very particular skillset which I haven't really exercised a lot since I graduated.

Never mind teaching the humanities, for which I think this is a genuine crisis, in class programming exams are basically the same thing as leetcode job interviews, and we all know what a bad proxy those are for "real" development work.

Also, just like how calculators are allowed in the exam halls, why not allow AI usage in exams? In real-life job you are not going to avoid use of calculator or AI. So why test people in a different context? I think the tests should focus on the skills in using calculator and AI.

My wife is a teacher. He school did this a long time ago, long before AI. But they also gave every kid a laptop and forced the teachers to move all tests/assignments to online applications with the curriculum picked out by the administrators (read as: some salesperson talked them into it). Even with assignments done in class, it's almost impossible to catch kids using AI when they're all on laptops all the time and she can't teach and monitor them all at the same time.

Bring back pencil and paper. Bring back calculators. Internet connected devices do not belong in the classroom.

I'm curious if we instead gave students an AI tool, but one that would intentionally throw in wrong things that the student had to catch. Instead of the student using LLMs, they would have one paid for by the school.

This is more brainstorming then a well thought-out idea, but I generally think "opposing AI" is doomed to fail. If we follow a montessori approach, kids are naturally inclined to want to learn thing, if students are trying to lie/cheat, we've already failed them by turning off their natural curiosity for something else.

Um. yea. This is the first time a non-deterministic technology has achieved mass adoption for every day use. Despite repeated warnings (which are not even close to the tenor of warnings they should broadcast), folks don’t understand that AI will likely hallucinate some or all of their answer.

A calculator will not, and even the closest aspect of buggy behavior for a calculator (exploring the fringes of floating point numbers, for example) is light years away from the hallucination of generated AI for general, every day questions.

The mass exuberance over generative AI has been clouding folks from the very real effects of over-adoption or AI, and we aren’t going to see the full impact of that for some time, and when we do, folks are going to ask questions like “how were we so dumb?” And of course the answer will be “no one saw this coming.”

My spouse is an educator with nearly 20 years in the industry, and even her school has adopted AI. It’s shocking how quickly it has taken hold, even in otherwise lagging adoption segments. Her school finally went “1-1” with devices in 2020, just prior to COVID.

So, in learning environments we might not have an option but to open the floodgates to AI use, but abandon most testing techniques that are not, more or less, pen and paper, in-person. Use AI as much as you want, but know that as a student you'll be answering tests armed only with your brain.

I do pity English teachers that have relied on essays to grade proficiency for hundreds of years. STEM fields has an easier way through this.

Also, all of these AI threats to public education can be mitigated if we just step 1-2 decades back and go the pen-and-paper way. I am yet to see any convincing argument in favor of digital/screen-based teaching methods being superior in any way than the traditional ones, on the contrary I have seen thousands of arguments against them.

So it is feasible (in principle) to give every student a different exam!

You’d use AI to generate lots of unique exams for your material, then ensure they’re all exactly the same difficulty (or extremely extremely close) by asking an LLM to reject any that are relatively too hard or too easy. Once you have generated enough individual exams, assign them to your students in your no-AI setting.

IS NO ONE GOING TO POINT OUT MULTIPLE OF THOSE DOODLES ARE WRONG???

One idea: Have students generate videos with their best "ELI5" explanations for things, or demos/teaching tools. Make the conciseness and clarity of the video video and the quality/originality of the teaching tools the grading criteria. Make the videos public, so classmates can compare themselves with their peers.

Students will be forced to learn the material and memorize it to make a good video. They'll be forced to understand it to create really good teaching tools. The public aspect will cause students to work harder not to feel foolish in front of their peers.

The beauty of this is that most kids these days want to be influencers, so they're likely to invest time into the assignment even if they're not interested in the subject.

It's simply too complex to fix. I think we'll see increased investment by corporates who do keep hiring on remediating the gaps in their workforce.

Most elite institutions will probably increase their efforts spent on interviewing including work trials. I think we're already seeing this with many of the elite institutions talking about judgment, emotional intelligence critical thinking as more important skills.

My worry is that hiring turns into a test of likeability rather than meritocracy (everyone is a personality hire when cognition is done by the machines)

Source: I'm trying to build a startup (Socratify) a bridge for upskilling from a flawed education system to the workforce for early stage professionals

It may not be obvious in a country with smaller student to teacher ratios, but for a place like India, you never have enough teachers for students.

Being able to provide courses, and homework digitally, reduced the amount of work required to grade and review work.

Then to add insult to injury, AI is removing entry level roles, removing other chances for people to do work which is easy to verify, practice and learn from.

Yes, yes, eventually tool use will result in increases in GDP. Except our incentives are not to hire more teachers, build more schools, and improve educational outcomes. Those are all public goods, not private goods. We aren’t going to tax firms further, because commerce must be protected, yet we will socialize the costs to society.