arXiv：cs.LG· Alexander Spangher, Sheldon Huang, Andreas Haupt, Noah D. Goodman, Diyi Yang, Daniel E. Ho, Sanmi Koyejo·· 3 小时前AI 评分60
arXiv 论文提出 CreativePreferences 数据集量化偏好的可验证与默会成分
Verifiable, Articulable, and Tacit Components of Preference
AI 导读
arXiv 论文 arXiv:2610.03025 发布 CreativePreferences 数据集，含 2.8M 文本、7 个创意领域、317M 条人类偏好判断和 42 个基准任务。
正文
Abstract:What makes a short story gripping; a news article newsworthy; or a math proof elegant? These constructs resist articulation or verification; their meaning is at least partially tacit. However, modern AI models are improved primarily via articulated constitutions, rubrics and verifiers (i.e. in RLAIF and RLVR); tacit components of preferences are typically understudied. We introduce a large, labeled preference dataset CreativePreferences, containing 2.8M texts labeled by 317M human preference judgments across 7 creative domains, with 42 benchmark tasks. We model these labels with executable programs, rubric banks and densely trained models (V, A and VAT, respectively). We observe robust articulability gaps, VAT-VA; and verifiability gaps, VAT-V; we estimate upper and lower bounds for each gap with a novel measurement approach that discovers articulable and verifiable metrics, identifies spurious variables and estimates the value of undiscovered metrics using capture-recapture. These gaps occur across all domains, even in domains traditionally treated as fully verifiable: correctness-centered domains (i.e. mathematics and software engineering) and claim- and novelty-centric domains (i.e. news, patents, peer review). The size of the gap varies based on domain (e.g. peer review and creative writing have the largest articulability gaps) and widens as more people take part in the judgment, consistent with Collins' collective tacit knowledge. We show two consequences: (1) on human generations, the full model more closely matches human preferences, often in disagreement with articulated criteria, and (2) in an analogy to Goodhart's law, articulating preference shifts it away from the tacit dimension. Articulability and verifiability gaps are consequential; we give recommendations on when tasks can be prompted; how learning mechanisms might improve; and when to leave judgments with humans.
| Comments: | 15 pages main text, 14 pages of references, 107-page appendix (136 pages total); 15 figures, 48 tables; 213 references |
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computers and Society (cs.CY); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.03025 [cs.AI] |
| (or arXiv:2610.03025v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.03025 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Alexander Spangher [view email]
[v1]
Fri, 2 Oct 2026 09:01:07 UTC (1,070 KB)
来源：arXiv：cs.LG · arxiv.org