Tag
Proposes an anytime-valid attribution method that uses a human-labeled anchor set and a betting e-process to distinguish whether score drift in LLM evaluation pipelines comes from the system or the judge, resolving the ambiguity caused by silent judge changes.
PACE introduces an anytime-valid commit gate for self-evolving agents that replaces greedy acceptance with a sequential hypothesis test, controlling false-commit probability and reducing churn while matching performance with lower variance.
Introduces Conformal Selective Acting (CSA), a deployment-time wrapper for RLVR-trained LLMs that provides anytime-valid selective risk control on individual streams, enabling safe deployment in regulated settings without pooling or long-run averages.
This paper introduces a statistical framework for adaptively auditing AI systems using Safe Anytime-Valid Inference (SAVI) to draw rigorous conclusions with limited data. It proposes a 'testing by betting' approach to validate model robustness while controlling type-I errors during adaptive sampling.