BOND distills Bayesian posteriors over six opponent priority orderings from a teacher LLM into an 8B student model, achieving Brier score 0.114 on CaSiNo while enabling posterior trajectory audits and outperforming a 70B baseline in calibration.
Judgment and Decision making , volume=
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
-
Distilling Bayesian Belief States into Language Models for Auditable Negotiation
BOND distills Bayesian posteriors over six opponent priority orderings from a teacher LLM into an 8B student model, achieving Brier score 0.114 on CaSiNo while enabling posterior trajectory audits and outperforming a 70B baseline in calibration.