arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.21368cs.LGcs.AIcs.CLstat.ML
收藏

通过自一致性采样和置信校准实现AI代理的黑盒可靠性认证

Black-Box Reliability Certification for AI Agents via Self-Consistency Sampling and Conformal Calibration

OPIT -- Open Institute of Technology(开放技术研究所)

Charafeddine Mouzouni

AI总结:

本文通过自一致性采样和置信校准方法,为AI系统提供可靠性评估,验证了其在不同任务和数据上的有效性,并展示了降低API成本的应用价值。

摘要:

给定一个黑盒AI系统和一个任务,从业者能以多大的置信度信任系统的输出?我们通过自一致性采样和置信校准得出一个可靠性水平——每个系统-任务对对应的一个数字,该数值作为黑盒部署的门禁,具有精确的有限样本、分布无关的保证。自一致性采样能指数级减少不确定性;置信校准保证在目标水平的1/(n+1)范围内正确,无论系统的错误如何——通过更大答案集对更难的问题进行透明可视化。较弱的模型获得较低的可靠性水平(不是准确性——见定义2.4):GPT-4.1在GSM8K上得分为94.6%,在TruthfulQA上得分为96.8%,而GPT-4.1-nano在GSM8K上得分为89.8%,在MMLU上得分为66.5%。我们验证了五个基准测试、五个来自三个家族的模型,以及合成和真实数据。可解决项目的条件覆盖超过0.93,在所有配置中;顺序停止将API成本降低约50%。

英文摘要:

Given a black-box AI system and a task, at what confidence level can a practitioner trust the system's output? We answer with a reliability level -- a single number per system-task pair, derived from self-consistency sampling and conformal calibration, that serves as a black-box deployment gate with exact, finite-sample, distribution-free guarantees. Self-consistency sampling reduces uncertainty exponentially; conformal calibration guarantees correctness within 1/(n+1) of the target level, regardless of the system's errors -- made transparently visible through larger answer sets for harder questions. Weaker models earn lower reliability levels (not accuracy -- see Definition 2.4): GPT-4.1 earns 94.6% on GSM8K and 96.8% on TruthfulQA, while GPT-4.1-nano earns 89.8% on GSM8K and 66.5% on MMLU. We validate across five benchmarks, five models from three families, and both synthetic and real data. Conditional coverage on solvable items exceeds 0.93 across all configurations; sequential stopping reduces API costs by around 50%.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑