对抗性概念搜索：从特征几何预测组合错误
Adversarial Concept Search: Predicting Compositional Errors From Feature Geometry
AI总结：
利用LLM的表征几何预测其组合失败模式，发现概念编码近正交时可靠组合，编码接近时因干扰导致失败，无需评估具体输入即可预测错误。
摘要：
人类并不总能直觉地判断哪些场景对LLM最具挑战性。为了捕捉具有挑战性的边缘案例，开发者要么设计对人类困难的问题，要么策划广泛的基准测试。如果我们能预先预测模型会在哪些场景上失败呢？在本文中，我们利用LLM的表征几何来预测它会在哪些概念组合上失败。我们将这种组合失败归因于显著特征之间的干扰。在需要系统组合的任务中——玩具程序化设置、多跳推理、多语言事实回忆——我们发现，当一对概念被编码为近似正交时，模型可靠地组合它们。当它们的线性编码接近时，产生干扰，模型无法组合它们。我们的方法可靠地预测了不同组合任务中的失败模式，无需评估特定输入。这些结果为利用表征几何识别高风险示例、构建有针对性的压力测试以及在现实部署中提供可扩展的主动学习基础奠定了基础。
英文摘要：
Humans cannot always intuit what scenarios are most challenging to LLMs. Hoping to capture challenging edge cases, developers either design problems to be difficult for humans or curate extensive benchmarks. What if we could instead anticipate which scenarios a model will fail on? In this paper, we use an LLM's representational geometry to predict which concept combinations it will fail on. We attribute this compositional failure to interference between salient features. In tasks that require systematic composition - toy programmatic settings, multihop reasoning, multilingual factual recall - we find that when a pair of concepts is encoded near-orthogonally, the model reliably composes them. When their linear encodings are close, producing interference, the model fails to compose them. Our method reliably anticipates failure modes across different compositional tasks, without evaluating specific inputs. These results lay the groundwork to use representational geometry to identify high-risk examples, construct targeted stress tests, and provide a scalable foundation for active learning in real-world deployment.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
