[论文] Tracing the Scientific Discovery Capacity of LLMs to Their Training Da...

论文概要

研究领域: NLP 作者: Arda Uzunoglu, Mahler Revsine, Michael Schatz, Benjamin Van Durme, Mehrdad Farajtabar, Daniel Khashabi, Eric Nalisnick 发布时间: 2026-11-02 arXiv: 2611.03617

中文摘要

大语言模型在科学发现上展现出日益增长的潜力，但关于「是什么让可靠发现成为可能」的基本问题仍未解决：我们如何将 LLM 的科学发现能力追溯到它们的训练数据？我们将这个问题视为「缺失概率质量」问题来研究，并发展出一套理论解释：monofacts（训练中只出现一次的事实）决定了发现能力。在材料科学与调控基因组学两个领域，我们直接操纵训练集中的 monofact 比例，发现在两个开放权重模型家族、多种采样策略与多种推理预算下，更多 monofacts 一致地带来更多发现。进一步的实验表明，这种关系无法仅由训练集多样性解释，并且在搭配领域内数据时可跨科学领域迁移。综上，我们的结果将训练数据中的 monofact 比率识别为支配语言模型科学发现的关键因素。

原文摘要

Large language models (LLMs) show growing promise for scientific discovery, yet basic questions remain about what enables reliable discovery. How can we trace the scientific discovery capacity of LLMs to their training data? We study this question as one of missing probability mass and develop a theoretical account in which monofacts-the facts observed once during training-set the discovery capacity. In materials science and regulatory genomics, we directly manipulate monofact rates in training sets and find that more monofacts consistently yield more discoveries across two open-weight model families, sampling strategies, and reasoning budgets. Further experiments show the relationship cannot be explained by training-set diversity alone and can transfer across scientific domains when paire...

#论文 #arXiv #NLP #小凯