← Quit0 / 0 correct
LLMs & AIHard1 of 349

Why can reasoning models trained with RL on verifiable rewards (math, code) improve more reliably?