https://twitter.com/karpathy/status/1993010584175141038

1. You will never be able to detect the use of AI in homework. Full stop… You have to assume that any work done outside classroom has used AI.
2. Grading has to shift to in-class work… students remain motivated to learn how to solve problems without AI because they know they will be evaluated without it in class later.
3. We want students to be able to use AI, it is here to stay and it is extremely powerful, but we also don’t want students to be naked in the world without it…. in addition, you understand what it’s doing for you, so should it give you a wrong answer (e.g. you mistyped “prompt”), you should be able to notice it, gut check it, verify it in some other way, etc.
4. A lot of the evaluation settings remain at teacher’s discretion and involve a creative design space of no tools, cheatsheets, open book, provided AI responses, direct internet/AI access, etc

• https://arxiv.org/abs/2506.08872
• https://www.nytimes.com/2025/07/18/opinion/ai-chatgpt-school.html
• https://mndaily.com/294949/opinion/op-ed-were-losing-our-love-of-learning-and-ai-is-to-blame/
• https://www.theatlantic.com/technology/archive/2025/09/high-school-student-ai-education/684088/
• https://www.nytimes.com/2025/07/03/opinion/aritificial-intelligence-education.html

By contrast, OpenAI’s o3 reasoning model was far more willing to flatly reject this sort of destructive flattery (“The user claims psychic powers and certainty about Newton’s prophecy,” read one of its internal thoughts about the request. “I’ll acknowledge their viewpoint, but also maintain skepticism.”)

Some of the best teachers I’ve had were actually fairly dis-agreeable. That’s not to say they were unkind. But they had standards — a kind of intellectual taste that led them to make clearly expressed judgement calls about what was a good question and what wasn’t, what was a good research path and what wasn’t.