arXiv：cs.LG· Soundouss Messoudi, Youness Mansar·· 3 小时前
System One 模型的校准与风险预算决策
Calibration and Risk-Budgeted Decisions with System One Models
AI 导读
研究针对 LLM 工作流中允许/阻止、丢弃/保留等二元决策，提出在不确定时默认采用保守回退，并控制被赋予效用最大化动作子集内的错误遗漏风险。在覆盖审核、安全与重排序的六个数据集上，朴素阈值选择基线在约半数设定中超出目标风险，而 Learn Then Test（LTT）为 7.8%。结果表明仅靠类概率接口不足以支撑部署决策。
正文
Abstract:LLM workflows and agents rely on a set of binary decisions like allow/block an input, or discard/retain a retrieved document. When human deferral is unavailable, uncertain cases must still receive an action. We study System One decision scores in this setting by defaulting to a conservative fallback when uncertain, and controlling the false omission risk within the subset of inputs assigned the utility-maximizing action. Across six datasets covering moderation, safety, and reranking, we compare general System One models with task baselines and a local open-weights fine-tuned System One model. We show that naive baselines for threshold selection exceed the target risk in roughly half of settings with defined test risk, compared with 7.8\% for Learn Then Test (LTT), which leverages finite-sample statistical guarantees. Our results show why relying on a probability-like interface is insufficient for deployment decisions and how risk-budgeted selection trades useful action against observed error.
| Subjects: | Machine Learning (stat.ML); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.19811 [stat.ML] |
| (or arXiv:2610.19811v1 [stat.ML] for this version) | |
| https://doi.org/10.48550/arXiv.2610.19811 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Soundouss Messoudi Dr [view email]
[v1]
Wed, 14 Oct 2026 08:54:01 UTC (254 KB)
来源：arXiv：cs.LG · arxiv.org