a16z crypto show · Crypto & Web3 · November 2026
The guests were debating whether AI-found mathematical results can outrun human understanding. Dan Boneh raises the case of a model settling P versus NP with a proof no person could read.
…humans, we can go to the machine and ask the machine to solve the problem we just raised, or we can try to solve it ourselves. And I claim that we will always kind of enjoy the process of solving for our own understanding, even though we know the machine can do it instantly and far faster than us.
Okay, but that's fine. But again, the way I think about that is that we're going to live in a world with a math truth oracle. So you can ask it any question you want, and it'll tell you, here's the answer.
Okay, so here's what scares me. Here's what scares me. Here's the thing that I'm really quite worried about. So let's say, let's pick one of our most important questions. Let's say P versus N P question. We would love to have a solution to a proof that P is not equal to N P. But what happens if the models actually come up with a solution? Except the solution is now a billion pages. Yeah, and no human will ever understand it. So what happens? I don't know what happens to mathematics when we
get to a point where right now we're nowhere near that. This is why I consider these as tools. They have all the
proofs they have produced,
we can quite easily understand. But what happens when we move beyond that and the proofs are just too long? Okay, let me