RLVR-trained LLMs show a stronger collapse in semantic branching entropy than in syntactic style entropy, suggesting the policy prunes genuinely distinct reasoning continuations.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
BODHI: Do LLMs Branch Out and Discover Heterogeneous Inferences?
RLVR-trained LLMs show a stronger collapse in semantic branching entropy than in syntactic style entropy, suggesting the policy prunes genuinely distinct reasoning continuations.