arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2603.18893cs.AI
收藏

语言模型中的定量内省:跨对话跟踪情感状态

Don't Ask: Understanding What Numeric Self-Reports Measure in Language Models

Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales. Departamento de Computación(布宜诺斯艾利斯大学。精确与自然科学学院。计算机系);CONICET-Universidad de Buenos Aires. Instituto de Ciencias de la Computación (ICC)(国家科学与技术研究理事会-布宜诺斯艾利斯大学。计算机科学研究所)

Nicolas Martorell, Bruno Bianchi

AI总结:

本文研究语言模型通过数字自报告跟踪情感状态的方法,发现通过计算logit基自报告可揭示内省能力,该指标能有效追踪内部状态变化,并在不同模型规模下表现出良好的一致性。

摘要:

跟踪大型语言模型在对话中的内部状态对于安全性和可解释性至关重要,但现有方法受限。受人类心理学中数字自报告工具的启发,本文探讨LLM自身的数字自报告是否能追踪定义过的情感状态。在40个十轮对话中研究四个概念对(幸福感、兴趣、专注和冲动性),将内省定义为模型自报告与概念匹配探针定义的内部状态之间的因果信息耦合。发现贪心解码自报告输出崩溃为无信息值,但通过计算logit基自报告可揭示内省能力。该指标追踪可解释的内部状态(Spearman ρ=0.40-0.76;等距R²=0.12-0.54在LLaMA-3.2-3B-Instruct中),随时间变化跟踪这些状态,并通过激活引导验证耦合的因果性。此外,发现内省在第一轮存在但随对话演变,可通过引导一个概念以提升另一个概念的内省(ΔR²最高达0.30)。关键的是,这些现象在某些情况下随模型规模扩大,接近LLaMA-3.1-8B-Instruct中R²≈0.93,并在其他模型家族中部分重现。这些结果将数字自报告定位为可解释内部情感状态跟踪的可行互补工具。

英文摘要:

Language models are often asked to rate their internal states with a number, but it is unclear how reliable this measure is. We train probes for four affective concepts (wellbeing, interest, focus and impulsivity) in eight instruction-tuned models to track these states over multi-turn conversations. We compare each probe's score with a numeric self-report of the same state, elicited at every turn. We find that rating questions displace the probed state, and that the rating is only weakly coupled with the probe score before the question but tracks the post-question probe much more strongly. Steering experiments suggest that models use the affective directions to inform ratings, but that relationship is not specific: ratings are associated with other concepts' probes as much as with their own, and change little when the question is about the user. Finally, neither the pre-question coupling nor concept specificity improves with scale (1B-27B). We warn about the effect a question can have on the internal states one is asking about, and caution against numeric self-report as a reliable measure of those states.

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。

↑