x
Research update: RL on Debate Games shows Proposal Accuracy uplift alongside Judge Hacking — LessWrong