[1718627440]

A system outputting correct facts, tells you nothing about the system's ability to prove correctness of facts. You can not assert that property of a system by treating it as a black box. If you are able to treat LLMs as a white box and prove correctness about their internal states, you should tell that to some very important people, that is an insight worth a lot of money.

[phendrenad2]

As usual, my argument brought all the people out of the woodwork who have some obsession about an argument that's tangential. Sorry to touch your tangent, bud.

[1718627440]

> LLMs not being able to detect correctness is just demonstrably false if you play around with LLM agents a bit.

How is telling you that this method of determining correctness is incapable of doing so, only tangential?

[phendrenad2]

Correctness and proven correctness are different things. I suspect you're a big Rocq Prover fan.