Vlad Tenev

being there, like the substrate existing for creating AI that's superhuman at solving mathematics.

But it's just, it wasn't there at the time.

It's still not quite there.

And I think we're harmonic, which is the company has a product called Aristotle.

Aristotle is very much at the frontier

And the big thing that is getting in the way, the big obstacle that we have to solve is the hallucinations.

So you can imagine if you're writing a history essay or just doing research, hallucinations are not that big of a deal.

I mean, annoying, but you can generally one hallucination, one mistake doesn't destroy the entire argument.

But if you have a complex mathematical theorem or something that you need to prove,

it could be thousands of steps and a hallucination in one of the steps essentially invalidates the whole thing so the whole thing is just garbage so in order to to solve mathematical super intelligence you have to solve this hallucination problem and make sure that every step of reasoning is logically sound and follows from the axioms and that you can trust the conclusions if you do that you solve

this subtraction being wrong problem.

Isn't that the case though with AI writing code?

It's 100 percent the case with AI writing code.

These things are very, very connected.

So the issue with vibe coding

is that you just create a lot of slop code.

You can think of this mountain of code.

Anytime you send another message saying, okay, add this feature, fix this bug, this thing screwed up, the code just increases, which I think works in some cases but it

creates this problem and changes the job of the software engineer from someone writing the code to someone trimming this bonsai tree or dealing with this conflagration of nonsense and clamping it down.

So there is a problem with vibe coding too, which is there's just hallucinations that you also have to solve.