Vladimir Tenev

but then it'll expand.

And I think as it expands, the use cases will get broader.

So beyond mathematicians, the next use case is creating software that's verified and correct.

And the first places where that's going to be useful is in mission-critical domains.

So domains where if your software screws up, the consequences are really huge.

So think chip design, right?

If you screw up the design of a chip and you're fabbing it, that costs hundreds of millions or billions.

And if you could speed that up and increase the...

uh the the reliability and you can verify that the design is good you might be able to like 5x the the speed of development there which is which is big talk about automotive control systems for for cars uh financial services and crypto how many times do you hear about a

blockchain vulnerability that leads to hundreds of millions of dollars being siphoned off of crypto protocols.

So I think it'll start there in verifiable correctness in safety critical domains.

But then we see a world where if the tools get good enough, the cost of

provably correct AI code will approach the cost of just any code.

They'll both go to zero.

And so then all software will be verifiably correct.

And then I think then it becomes much more broadly applicable.

At first, yeah, I think it's mostly enterprise.

But I think there is a path where if we actually succeed in this, then you could have something like a chat GPT or a clod code.

Aristotle now feels very much like clod code.

You could have one where you're very, very highly confident that what it's telling you is right.