PerMix-RLVR balances persona robustness and fidelity in RLVR-trained LLMs, delivering +21.2% PSS on MATH500 and +11.4% fidelity on PersonaGym.
It’s like a secret code that we can’t break
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
-
PerMix-RLVR: Preserving Persona Expressivity under Verifiable-Reward Alignment
PerMix-RLVR balances persona robustness and fidelity in RLVR-trained LLMs, delivering +21.2% PSS on MATH500 and +11.4% fidelity on PersonaGym.