A major bottleneck in characterizing the failure modes of generative AI systems is the cost and time of annotation and evaluation. Consequently, adaptive testing paradigms have gained popularity, where one opportunistically decides which cases and how many to annotate based on past results. While this framework is highly practical, its extreme flexibility makes it difficult to draw statistically rigorous conclusions, as it violates classical assumptions: the number of observations is typically limited (often 10 to 50 cases) and decisions regarding sampling and stopping are made in the midst of data collection rather than based a pre-specified rule. To characterize what statistical inferences can be drawn from highly adaptive audits, we introduce a hypothesis testing framework from two "dueling" perspectives: (i) the model's null that asserts there is no failure mode with performance below a target threshold versus (ii) the auditor's null that asserts they have a sampling strategy that will uncover a failure mode. Leveraging Safe Anytime-Valid Inference (SAVI), we formalize the auditor as conducting "testing by betting," which translates into simultaneous e-processes for testing the dueling null hypotheses. Furthermore, if the auditor is sufficiently powerful, we prove that these two hypotheses are asymptotically inverses of each other, in that passage of such an audit certifies the AI system as being globally robust. Empirically, we demonstrate that our proposed testing procedures maintain anytime-valid Type-I error control, outperform pre-specified testing methods, and can reach statistically rigorous conclusions sometimes with as few as 20 observations.
Predicting fixed-sample test decisions enables anytime-valid inference
AI & Robotics Institute, Ellison Institute of Technology(艾利森技术学院人工智能与机器人研究所);Department of Statistics, University of Oxford(牛津大学统计系);University of Kent(肯特大学)
CITE: Anytime-Valid Statistical Inference in LLM Self-Consistency
Komaba Institute for Science, Graduate School of Arts and Sciences, The University of Tokyo(东京大学艺术科学研究生院Komaba研究所);Nagoya University(名古屋大学);NARA Institute of Science and Technology (NAIST) / Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)(NAIST科学与技术研究所 / 摩洛哥本·泽德人工智能大学);Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) / RIKEN AIP(摩洛哥本·泽德人工智能大学 / RIKEN AIP);Graduate School of Arts and Sciences, The University of Tokyo / RIKEN AIP / Kyoto University(东京大学艺术科学研究生院 / RIKEN AIP / 京都大学)
Division of Systems and Control, Department of Information Technology, Uppsala University, Sweden(信息科技系系统与控制 division,乌普萨拉大学,瑞典);Science for Life Laboratory, Uppsala, Sweden(生命科学实验室,乌普萨拉,瑞典)
Anytime-Valid Answer Sufficiency Certificates for LLM Generation via Sequential Information Lift
Department of Computer Science, Iowa State University(计算机科学系,爱荷华州立大学);Department of Civil, Construction & Environmental Engineering, Iowa State University(土木、建设与环境工程系,爱荷华州立大学)