arXiv cs.AI· Marilyn Zhang, Tianfeng Chen, Mark E. Whiting·· 13 小时前AI 评分42
生物学上下文并不能可靠提升 LLM 引导的实验优化效果
Biological Context Does Not Reliably Improve LLM-Guided Experimental Optimization
AI 导读
在 45 个生物学任务上测试 19 个 LLM 发现，去掉生物学含义的域无关提示在 58.7% 的任务轨迹和 61.9% 的终点指标上优于域感知提示。该失败模式为生物学特有，在同样构建的教育任务上结果反转；域感知提示让模型初始设计更好，但对反馈的响应更差，部分原因是锚定效应。将 Llama 3.1 8B 后训练至前沿级原始优化能力，并不能迁移为生物学感知优化能力的提升。
正文 · 原文
Abstract:Autonomous labs increasingly deploy large language models (LLMs) as closedloop biology optimizers that propose a design, read the outcome, then propose again. This is partly because of an assumption that biological knowledge helps an LLM choose better designs than Bayesian optimization, which has no access to biological context. We test that assumption on 45 biology tasks reconstructed from 2,485 experimental rows across 341 published papers, scored by oracles fitted to those rows, each run over 30 iterations under two prompt conditions. The domain-aware condition provides the published parameter names, units, and labels, and the domain-agnostic renames them so they are not biologically meaningful, with search space and feedback identical. Across 19 LLMs domain-agnostic outperforms domain-aware on 58.7% of tasks by trajectory and 61.9% by endpoint. The failure is specific to biology, reversing on education tasks built the same way. Domain-aware prompting tends to help models propose better initial designs, while making them respond worse to feedback. Part of the reason is anchoring, with models keeping the conventional categorical choice even when the oracle scores a different (published best) value higher. Post-training Llama 3.1 8B to frontier-level raw optimization does not transfer to biology-aware optimization improvements. Biology-aware optimization therefore looks like a separate capability that autonomous laboratory loops should target directly.
| Comments: | NeurIPS 2026 Workshop: I Can't Believe It's Not Better (ICBINB): Failure Modes of AI in Biology |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| ACM classes: | I.2.1 |
| Cite as: | arXiv:2605.15341 [cs.LG] |
| (or arXiv:2605.15341v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2605.15341 arXiv-issued DOI via DataCite |
Submission history
From: Marilyn Zhang [view email]
[v1]
Thu, 14 May 2026 19:10:45 UTC (303 KB)
[v2]
Tue, 20 Oct 2026 02:25:59 UTC (52 KB)
来源：arXiv cs.AI · arxiv.org