Likes Zelda. Likes Metroid.

Transcript context

…Some of the providers are starting to work on models for education designed to not give… actually, I haven’t used them, but I would guess they’re designed to not give all the information at once— —and make people work for it. I think you could train models to do this and it would be a wonderful contribution. In the book, you had to reevaluate every decision— Which is such a great example. I think there’s a chance we work on it at AI2, which I think would be so fun. It makes sense. I did something like that the other day for video games. In my spare time, I like video games with puzzles, like Zelda and Metroid. There’s this new game where I got really stuck. I didn’t want to struggle for two days, so I used an LLM. But I told it, “Please don’t add any spoilers. I’m at this point; what do I have to do next?” You can do the same thing for math where you say, “I’m at this point and I’m getting stuck. Don’t give me the full solution, but what is something I could try?” You kind of carefully probe it. But the problem is that it requires discipline. A lot of people enjoy math, but there are also a lot of people who need to do it for their homework, and then it’s just a shortcut. We can develop an educational LLM, but the other LLMs are still there, and there’s still a temptation to use them. I think a lot of people, especially in college, understand the stuff they’re passionate about— —they’re self-aware about it, and they understand it shouldn’t be easy. Like, I think we just have to develop a good taste— —talk about research taste, like school taste about stuff that you should be struggling on— —and stuff you shouldn’t be struggling on. Which is tricky to know, because sometimes you don’t have good long-term vision about what would be actually useful to you in your career. But you have to develop that taste, yeah.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.