How to learn with AI without letting it do the thinking for you
Two habits come from Nobel physicist John Martinis on studying with AI: don't look up the answers to your practice problems, and cross-check one AI against a different one, the way he does. The other three are things the show's guests said that are worth remembering when a chatbot can hand you an answer — Martinis deriving physics from a few basics instead of memorizing, his point that getting good takes time and persistence, and Vineet Buch's that the hard part is often finding the right question, not the answer. Each starts from a clip you can watch; any line that begins 'Applied to AI' is the editor's framing.
Co-founder & CEO, ClimateAi · WEF Young Global Leader · Forbes 30 Under 30 · Stanford
The short answer
- Don't look up the answers to your practice problems. Working through them is the learning; hand them in solved and you skipped that.
- Martinis checks an AI answer against a different AI; consistency is a signal, not a guarantee.
- Martinis learned physics by holding a few basics and deriving the rest, not memorizing.
- Finding the right question, not the answer, is the hard part. Vineet Buch says school never taught it.
Should you use AI to solve your problem sets?
Not for practice problems. Martinis's point is that a problem set exists to teach you to solve problems, so the moment you look up the answer you can hand it in but you have failed to build the skill it was there to build. When you are truly stuck after real effort, he suggests fellow students, or a professor or TA who nudges you toward the answer instead of handing it over. Applied to AI: the chatbot is the fastest way to look it up, which is exactly why it is the wrong first move on a practice problem.
“as soon as you look it up, you can turn in your problem set, but you kind of have failed in your ability to like really understand how to learn problems.”Watch John Martinis say this at 1:00:45
How do you use AI to learn without being misled by it?
Check it against a different AI. Martinis, who uses AI constantly in his own research, runs the same question through two different AIs to see whether they are consistent. Applied to AI: consistency is a useful signal, not a guarantee, since two systems can agree and both be wrong, so the thing to avoid is trusting a single unchecked answer.
“I generally use two AIs to see if the results are consistent with each other.”Watch John Martinis say this at 40:12
How did Martinis learn physics with a bad memory?
He made it an advantage. In physics, he says, there are only a few basic concepts and formulas to hold, and you derive the rest, so a weak memory pushed him to understand and rebuild ideas rather than store them.
“I have a bad memory. But this is in fact really good for physics because in physics, have a few basic concepts and formulas to memorize. And then you tend to derive everything.”Watch John Martinis say this at 00:41
Is there a shortcut to getting good, or do you put in the time?
No shortcut. Martinis's rule for getting good at most things is plain: put in the time and the persistence and you get there, though he adds that it may make you good without making you the best.
“there's this concept that you can get really good at something if you just put in time and persistence and do it.”Watch John Martinis say this at 11:46
What's the hard part that school doesn't teach?
Asking the right question. Vineet Buch, who topped the IIT entrance exam, says that in real life the hard part is not finding the answer but finding the right question, and that school does not train you for it.
“In real life, the hard part is not finding the answer, it is finding the right question. And academia does not train you for that.”Watch Vineet Buch say this at 22:56
Each block starts from something a guest actually said, with a clip you can watch. Any line that begins 'Applied to AI' is the editor's framing, not the guest's words. Written by Himanshu Gupta.