Prover-Verifier Games improve legibility of LLM outputs

Basically GAN but for LLMs? Neat

Very interesting. I always see the two techniques as complementary.

(openAI) Previous discussion:

https://news.ycombinator.com/item?id=40988076