Calibrating the Confidence of Large Language Models by Eliciting Fidelity
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909341113647104 |
|---|---|
| author | Zhang, Mozhi Huang, Mianqiu Shi, Rundong Guo, Linsen Peng, Chong Yan, Peng Zhou, Yaqian Qiu, Xipeng |
| author_facet | Zhang, Mozhi Huang, Mianqiu Shi, Rundong Guo, Linsen Peng, Chong Yan, Peng Zhou, Yaqian Qiu, Xipeng |
| contents | Large language models optimized with techniques like RLHF have achieved good alignment in being helpful and harmless. However, post-alignment, these language models often exhibit overconfidence, where the expressed confidence does not accurately calibrate with their correctness rate. In this paper, we decompose the language model confidence into the \textit{Uncertainty} about the question and the \textit{Fidelity} to the answer generated by language models. Then, we propose a plug-and-play method to estimate the confidence of language models. Our method has shown good calibration performance by conducting experiments with 6 RLHF-LMs on four MCQA datasets. Moreover, we propose two novel metrics, IPR and CE, to evaluate the calibration of the model, and we have conducted a detailed discussion on \textit{Truly Well-Calibrated Confidence}. Our method could serve as a strong baseline, and we hope that this work will provide some insights into the model confidence calibration. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2404_02655 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Calibrating the Confidence of Large Language Models by Eliciting Fidelity Zhang, Mozhi Huang, Mianqiu Shi, Rundong Guo, Linsen Peng, Chong Yan, Peng Zhou, Yaqian Qiu, Xipeng Computation and Language Large language models optimized with techniques like RLHF have achieved good alignment in being helpful and harmless. However, post-alignment, these language models often exhibit overconfidence, where the expressed confidence does not accurately calibrate with their correctness rate. In this paper, we decompose the language model confidence into the \textit{Uncertainty} about the question and the \textit{Fidelity} to the answer generated by language models. Then, we propose a plug-and-play method to estimate the confidence of language models. Our method has shown good calibration performance by conducting experiments with 6 RLHF-LMs on four MCQA datasets. Moreover, we propose two novel metrics, IPR and CE, to evaluate the calibration of the model, and we have conducted a detailed discussion on \textit{Truly Well-Calibrated Confidence}. Our method could serve as a strong baseline, and we hope that this work will provide some insights into the model confidence calibration. |
| title | Calibrating the Confidence of Large Language Models by Eliciting Fidelity |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2404.02655 |