MaR is a metacognition-inspired RL method that rewards LLMs on task knowledge, regulation fidelity, and answer correctness, reporting gains up to 11% over baselines on 22 benchmarks.
Do not set s = 1 merely because the response is brief or lacks one labeled section
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Metacognition as Reward: Reinforcing LLM Reasoning via Knowledge and Regulation Signals
MaR is a metacognition-inspired RL method that rewards LLMs on task knowledge, regulation fidelity, and answer correctness, reporting gains up to 11% over baselines on 22 benchmarks.