Saved in:
Bibliographic Details
Main Authors: Zhang, Qingjie, Wang, Di, Qian, Haoting, Li, Yiming, Zhang, Tianwei, Huang, Minlie, Xu, Ke, Li, Hewu, Liu, Yan, Qiu, Han
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2412.14959
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909689650872320
author Zhang, Qingjie
Wang, Di
Qian, Haoting
Li, Yiming
Zhang, Tianwei
Huang, Minlie
Xu, Ke
Li, Hewu
Liu, Yan
Qiu, Han
author_facet Zhang, Qingjie
Wang, Di
Qian, Haoting
Li, Yiming
Zhang, Tianwei
Huang, Minlie
Xu, Ke
Li, Hewu
Liu, Yan
Qiu, Han
contents Intrinsic self-correction was proposed to improve LLMs' responses via feedback prompts solely based on their inherent capability. However, recent works show that LLMs' intrinsic self-correction fails without oracle labels as feedback prompts. In this paper, we aim to interpret LLMs' intrinsic self-correction for different tasks, especially for those failure cases. By including one simple task and three complex tasks with state-of-the-art (SOTA) LLMs like ChatGPT families (o1, 4o, 3.5-turbo) and Llama families (2-7B, 3-8B, and 3.1-8B), we design three interpretation methods to reveal the dark side of LLMs' intrinsic self-correction. We identify intrinsic self-correction can (1) cause LLMs to waver both intermedia and final answers and lead to prompt bias on simple factual questions; (2) introduce human-like cognitive bias on complex tasks. In light of our findings, we also provide two simple yet effective strategies for alleviation: question repeating and supervised fine-tuning with a few samples. We open-source our work at https://x-isc.info/.
format Preprint
id arxiv_https___arxiv_org_abs_2412_14959
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Understanding the Dark Side of LLMs' Intrinsic Self-Correction
Zhang, Qingjie
Wang, Di
Qian, Haoting
Li, Yiming
Zhang, Tianwei
Huang, Minlie
Xu, Ke
Li, Hewu
Liu, Yan
Qiu, Han
Computation and Language
Intrinsic self-correction was proposed to improve LLMs' responses via feedback prompts solely based on their inherent capability. However, recent works show that LLMs' intrinsic self-correction fails without oracle labels as feedback prompts. In this paper, we aim to interpret LLMs' intrinsic self-correction for different tasks, especially for those failure cases. By including one simple task and three complex tasks with state-of-the-art (SOTA) LLMs like ChatGPT families (o1, 4o, 3.5-turbo) and Llama families (2-7B, 3-8B, and 3.1-8B), we design three interpretation methods to reveal the dark side of LLMs' intrinsic self-correction. We identify intrinsic self-correction can (1) cause LLMs to waver both intermedia and final answers and lead to prompt bias on simple factual questions; (2) introduce human-like cognitive bias on complex tasks. In light of our findings, we also provide two simple yet effective strategies for alleviation: question repeating and supervised fine-tuning with a few samples. We open-source our work at https://x-isc.info/.
title Understanding the Dark Side of LLMs' Intrinsic Self-Correction
topic Computation and Language
url https://arxiv.org/abs/2412.14959