Tampered ML：评估 LLM 对机器学习研究的判断力
Tampered ML: Evaluating LLM Judgment of Machine Learning Research
AI 导读
一项用于自动对齐研究的基准 Tampered ML 发布，包含 1,858 对实验提案，每对由已发表机器学习论文重构的提案和一个引入针对性方法论缺陷的最小改动反事实组成，模型需在两种呈现顺序中找出方法学上更可靠的提案。
正文 · 原文
Abstract:One approach to accelerating progress on AI alignment is to automate parts of alignment research itself. Doing so requires systems that can perform reliably on hard-to-supervise research tasks, including identifying flaws in research proposals and evaluating whether a proposed experiment supports its intended claim. To evaluate this capability, we introduce Tampered ML, a benchmark of 1,858 item-pairs, each consisting of an experiment proposal reconstructed from a published machine learning paper and a minimally edited counterfactual that introduces a targeted methodological flaw. Models must identify the more methodologically sound proposal in both presentation orders. A human validation study on a stratified sample estimates that 86% of items are valid (95% CI: 81-91%). Frontier models perform above chance but remain far from reliable (with Opus 5 achieving 77.9% accuracy). They are highly sensitive to the order of presentation and favor flawed designs that appear more specific, transparent, or conventional.
| Comments: | 28 pages, 10 figures. Accepted at EMNLP 2026 (Main Conference) |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Digital Libraries (cs.DL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.32114 [cs.CL] |
| (or arXiv:2610.32114v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.32114 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Soroush Ebadian [view email]
[v1]
Fri, 23 Oct 2026 03:32:38 UTC (2,023 KB)
来源：arXiv cs.AI · arxiv.org