PROVER-VERIFIER IMPROVES LLM LEGIBILITY

Summary from the TLDR AI newsletter

OpenAI trained a strong model to output more legible text for a weak model to grade accurately and found that this led to general purpose improvements in overall legibility.