arXiv cs.AI· Jianlin Chen, Wenhui Chen, Chi Man Vong·· 3 天前AI 评分22
基于目录的 LLM 优化循环中已拒绝候选为何被重新提出
Visible but Not Avoided: State-Dependent Re-Proposal of Rejected Candidates in Catalog-Based LLM Optimization Loops
AI 导读
在三个开源 LLM 通过从 40 项编辑中命名一项来改进 GUI 智能体的目录式优化循环中，干净基态探针会低估提出者自身循环中的重复提出率（循环减探针的宏平均率为 0.32 [0.27, 0.37]），已评估的 harness-edit 配对提议占用 28-63% 的有效提出迭代。
正文 · 原文
Abstract:LLM optimization loops often show the proposer the candidates it has already tried and rejected, but leave it to the proposer to avoid them. We ask whether avoidance measured in a static probe predicts avoidance in the states an optimizer's own trajectory produces. In catalog-based loops where three open LLMs improve a GUI agent by naming one of 40 edits, a re-proposal is identified exactly by the edit's name. Clean base-state probes underpredict re-proposal in the proposers' own loops (macro-averaged loop-minus-probe rate 0.32 [0.27, 0.37]), and proposals of already evaluated harness-edit pairs consume 28-63% of valid proposer iterations. The two settings reject different edits: a probe's rejected edits are mostly ones the proposer did not choose, whereas a loop's are its own earlier choices, which remain preferred when the history is removed. Ordinary history lowers re-proposal for two proposers but raises it for the third, and the effects of failure evidence are model-specific. Stating the rejection status in one sentence lowers re-proposal in every model (-0.16 [-0.22, -0.09] relative to a neutral list of the same IDs). For catalog-based loops with exact edit identities, rejection avoidance should be measured on the rejected sets the loop produces.
| Comments: | 22 pages, 6 figures |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.15738 [cs.AI] |
| (or arXiv:2610.15738v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.15738 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jianlin Chen [view email]
[v1]
Sun, 11 Oct 2026 16:18:11 UTC (139 KB)
来源：arXiv cs.AI · arxiv.org