9 pointsby jonnonz8 hours ago1 comment
  • thx674 hours ago
    > Since monitoring task performance and adapting behavior accordingly are central to metacognition, we posit that mod- els capable of accurately judging their own performance are better positioned to improve it. We operationalize this idea via two novel mechanisms: reinforcement learning with metacognitive feedback (RLMF), a paradigm to refine completion rankings during preference optimization based on the quality of a model’s self- judgments of performance, and metacognitive data selection, which uses similar self-judgments to identify high-value training examples, outperforming naive active learning.