PulseAugur
实时 12:59:59
English(EN) What LLMs explain is not what they believe: Evaluating explanation sufficiency under models' own input beliefs

新指标 SCSuff 评估 LLM 解释的充分性

一项新的研究论文介绍了一种名为 SCSuff 的信息论指标,用于评估大型语言模型(LLMs)生成的自由文本解释的充分性。该研究提出,解释的充分性可能依赖于分布,并建议使用 LLM 本身来生成替代输入,从而捕捉其信念。实验表明,LLM 的解释普遍不足,并且与模型大小或准确性之间的相关性较弱,尽管 SCSuff 分数可以从模型的内部表示中预测出来。 AI

影响 这项研究可能带来更可靠、更值得信赖的 LLM 解释,这对于高风险应用至关重要。

排序理由 介绍评估 LLM 解释新指标的研究论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv stat.ML 阅读 →

AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →

新指标 SCSuff 评估 LLM 解释的充分性

报道来源 [2]

  1. arXiv stat.ML TIER_1 English(EN) · Nhi Nguyen, Shauli Ravfogel, Rajesh Ranganath ·

    What LLMs explain is not what they believe: Evaluating explanation sufficiency under models' own input beliefs

    arXiv:2606.28615v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly deployed in high-stakes domains, where free-text explanations such as chain-of-thought and post-hoc rationales are used to justify model outputs. Yet it remains unclear whether these e…

  2. arXiv stat.ML TIER_1 English(EN) · Rajesh Ranganath ·

    What LLMs explain is not what they believe: Evaluating explanation sufficiency under models' own input beliefs

    Large language models (LLMs) are increasingly deployed in high-stakes domains, where free-text explanations such as chain-of-thought and post-hoc rationales are used to justify model outputs. Yet it remains unclear whether these explanations are sufficient, i.e., if they contain …