Lookers-On See Most of the Game: An External Insight-Guided Method for Enhancing Uncertainty Estimation
Keywords: Large Language Models, Uncertainty Estimation, Trusty AI
TL;DR: We propose a external insight-driven method to enhance uncertainty estimation by integrating a lightweight model trained on global information.
Abstract: Large Language Models (LLMs) have gained increasing attention for their impressive capabilities, alongside concerns about the reliability arising from their potential to generate hallucinations and factual inaccuracies. Uncertainty estimation for LLMs aims to quantify the uncertainty of model outputs, where high uncertainty scores indicate potential errors, signaling the need for rejection or further evaluation. However, existing methods often limited by inherent biases of LLMs like over-confidence and under-confidence. In this paper, we propose an external insight-driven correction method for refining uncertainty estimation. This method integrates uncertainty scores derived from a lightweight model trained on global information with those from existing uncertainty estimation approaches, providing a more robust solution. We present comprehensive experimental results that demonstrate the effectiveness and generalizability of our method across various models, datasets, and consistently surpassing all baselines.
Supplementary Material: zip
Primary Area: foundation or frontier models, including LLMs
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 13250
Loading