大模型越大会更自信地犯错，暴露推理缺陷。
Pitfalls of Scale: Investigating the Inverse Task of Redefinition in Large Language Models
- 让大模型重新定义物理常数，测试其推理能力。
- 模型规模越大，错误率越高，但自信心反而更强。
- 即使优化提示方式，模型仍会死记硬背旧值。
随着大型语言模型（LLMs）规模增大，反向任务可揭示潜在的推理漏洞。本文研究重定义任务：为已知物理常数和单位赋予新值，观察模型是否能相应调整回答。结果显示，模型性能随规模增长而下降，同时错误判断的自信心上升。尽管提示策略或输出格式等因素有一定影响，但模型仍倾向于依赖记忆中的原始数值，难以真正理解概念重定义的本质。
原文摘要 · Abstract (English)
Inverse tasks can uncover potential reasoning gaps as Large Language Models (LLMs) scale up. In this work, we explore the redefinition task, in which we assign alternative values to well-known physical constants and units of measure, prompting LLMs to respond accordingly. Our findings show that not only does model performance degrade with scale, but its false confidence also rises. Moreover, while factors such as prompting strategies or response formatting are influential, they do not preclude LLMs from anchoring to memorized values.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。