强化学习
突破级
暂无讲解视频
收录解读
Your Language Model is Its Own Critic: Reinforcement Learning with Value Estimation from Actor's Internal States 关注的是一个可复用的 AI 系统或评测问题，而不是单点 demo。
POISE estimates RLVR baselines from policy internal states with negligible extra critic cost.
It is a reusable RLVR efficiency primitive, replacing large external critics or many rollouts with online internal-state value estimation.
它没有更高，是因为这些新 arXiv 工作仍需要更多独立复现、真实系统部署和长期社区采用来确认影响。