RL on binary rewards boosts LLM factual recall by ~27% relative across models by redistributing probability mass to latent correct answers rather than acquiring new knowledge.
Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou
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
-
Beyond Reasoning: Reinforcement Learning Unlocks Parametric Knowledge in LLMs
RL on binary rewards boosts LLM factual recall by ~27% relative across models by redistributing probability mass to latent correct answers rather than acquiring new knowledge.