微调大语言模型让它们知道自己知道什么：ESMA 方法被 NeurIPS 2026 接收
Fine-Tuning Language Models to Know What They Know
AI 导读
一篇被 NeurIPS 2026 接收的论文提出衡量并提升 LLM 元认知能力的框架，用 d'_type2 指标在排除偏差和启发式干扰的情况下分离元认知能力。论文提出用于元认知对齐的进化策略 ESMA，在未见过的数据集、语言和新知识上展现出稳健泛化。渐进参数更新显示这些提升由一组稀疏参数驱动，为定向元认知优化提供了潜在路径。
正文 · 原文
Abstract:Evaluating true metacognition in Large Language Models (LLMs) is difficult due to biases and heuristics. This paper presents a framework to measure and enhance LLM metacognition while controlling for these biases. A measurement method using the $d'_{\rm type2}$ metric is established to isolate metacognitive ability. The Evolution Strategy for Metacognitive Alignment (ESMA) is proposed, demonstrating robust generalization across unseen datasets, languages, and newly acquired knowledge. Gradual parameter updates reveal that these improvements are driven by a sparse set of parameters, opening a potential avenue for targeted metacognitive optimization.
| Comments: | Accepted to NeurIPS 2026 |
| Subjects: | Neural and Evolutionary Computing (cs.NE); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Neurons and Cognition (q-bio.NC) |
| Cite as: | arXiv:2602.02605 [cs.NE] |
| (or arXiv:2602.02605v3 [cs.NE] for this version) | |
| https://doi.org/10.48550/arXiv.2602.02605 arXiv-issued DOI via DataCite |
Submission history
From: Sangjun Park [view email]
[v1]
Mon, 2 Feb 2026 04:08:13 UTC (1,441 KB)
[v2]
Sun, 24 May 2026 06:55:56 UTC (1,449 KB)
[v3]
Tue, 10 Nov 2026 20:13:08 UTC (1,462 KB)
来源：arXiv cs.AI · arxiv.org