对话模型对自己的回答过度自信，新方法可有效纠正。
Large Language Models Are Overconfident in Their Own Responses
- 将模型回复伪装成用户输入，缓解自身回答的过度自信
- 实测模型对自己回答的置信度高出26%
- 无需重训练，适合部署时快速提升可靠性
先前研究发现，经过指令微调的大语言模型（LLM）校准性不如基础预训练模型。但对话模板对对话式LLM校准性的影响尚不明确。本文通过解耦后训练算法与对话格式的影响，发现尽管指令微调本身会损害校准性，而对话模板通过‘所有权偏差’进一步加剧问题——模型对其自身回答的置信度显著高于对相同内容由用户提供的回答。在六种近期开源大模型、三个基准测试和三种置信度获取方法上的实验表明，模型对自己的回答置信度最高可高出26%。基于此，我们提出一种简单推理阶段策略：在置信度评估时将模型回答视为用户输入。该方法无需重新训练，即可使过自信现象显著缓解，校准性能提升达26%，缩小了基础模型与指令微调模型之间的差距。
原文摘要 · Abstract (English)
Prior work has shown that instruction-tuned large language models (LLMs) are less well calibrated than their base pre-trained counterparts. However, little is known about the frequently used chat template's effect on the calibration of conversational LLMs. In this work, we investigate the mechanisms driving this miscalibration by decoupling the effects of the post-training algorithm and the chat format. We find that, while instruction tuning fundamentally harms calibration, the chat template aggravates the issue through an "ownership bias" -- models are significantly more confident in their own answers than in identical answers provided by a user. Extensive experiments across six recent open-weight LLMs, three benchmarks, and three confidence elicitation methods show that models assign up to 26% higher confidence to their own responses. Leveraging this insight, we propose a simple inference-time strategy: framing the model's answer as user input during confidence elicitation. This approach significantly reduces overconfidence and improves calibration by up to 26% without the need for retraining, narrowing the gap between base and instruction-tuned models.
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。