See What LLMs Cannot Answer: A Self-Challenge Framework for Uncovering LLM Weaknesses

• URL: http://arxiv.org/abs/2408.08978v2
• Date: Tue, 1 Oct 2024 01:40:14 GMT
• Title: See What LLMs Cannot Answer: A Self-Challenge Framework for Uncovering LLM Weaknesses
• Authors: Yulong Chen, Yang Liu, Jianhao Yan, Xuefeng Bai, Ming Zhong, Yinghao Yang, Ziyi Yang, Chenguang Zhu, Yue Zhang,
• Abstract summary: We propose a Self-Challenge evaluation framework with human-in-the-loop. Starting from seed instances that GPT-4 fails to answer, we prompt GPT-4 to summarize error patterns that can be used to generate new instances. We then build a benchmark, SC-G4, consisting of 1,835 instances generated by GPT-4 using these patterns, with human-annotated gold responses.
• Score: 51.975495361024606
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The impressive performance of Large Language Models (LLMs) has consistently surpassed numerous human-designed benchmarks, presenting new challenges in assessing the shortcomings of LLMs. Designing tasks and finding LLMs' limitations are becoming increasingly important. In this paper, we investigate the question of whether an LLM can discover its own limitations from the errors it makes. To this end, we propose a Self-Challenge evaluation framework with human-in-the-loop. Starting from seed instances that GPT-4 fails to answer, we prompt GPT-4 to summarize error patterns that can be used to generate new instances and incorporate human feedback on them to refine these patterns for generating more challenging data, iteratively. We end up with 8 diverse patterns, such as text manipulation and questions with assumptions. We then build a benchmark, SC-G4, consisting of 1,835 instances generated by GPT-4 using these patterns, with human-annotated gold responses. The SC-G4 serves as a challenging benchmark that allows for a detailed assessment of LLMs' abilities. Our results show that only 44.96\% of instances in SC-G4 can be answered correctly by GPT-4. Interestingly, our pilot study indicates that these error patterns also challenge other LLMs, such as Claude-3 and Llama-3, and cannot be fully resolved through fine-tuning. Our work takes the first step to demonstrate that LLMs can autonomously identify their inherent flaws and provide insights for future dynamic and automatic evaluation.

Related papers

• LLM Self-Correction with DeCRIM: Decompose, Critique, and Refine for Enhanced Following of Instructions with Multiple Constraints [86.59857711385833]
  We introduce RealInstruct, the first benchmark designed to evaluate LLMs' ability to follow real-world multi-constrained instructions. To address the performance gap between open-source and proprietary models, we propose the Decompose, Critique and Refine (DeCRIM) self-correction pipeline. Our results show that DeCRIM improves Mistral's performance by 7.3% on RealInstruct and 8.0% on IFEval even with weak feedback.
  arXiv  Detail & Related papers  (2024-10-09T01:25:10Z)
• Can LLMs replace Neil deGrasse Tyson? Evaluating the Reliability of LLMs as Science Communicators [22.567933207841968]
  Large Language Models (LLMs) and AI assistants are experiencing exponential growth in usage among both expert and amateur users. In this work, we focus on evaluating the reliability of current LLMs as science communicators. We introduce a novel dataset, SCiPS-QA, comprising 742 Yes/No queries embedded in complex scientific concepts.
  arXiv  Detail & Related papers  (2024-09-21T06:48:32Z)
• Navigating the Labyrinth: Evaluating and Enhancing LLMs' Ability to Reason About Search Problems [59.72548591120689]
  We introduce a new benchmark, SearchBench, containing 11 unique search problem types. We show that even the most advanced LLMs fail to solve these problems end-to-end in text. Instructing LLMs to generate code that solves the problem helps, but only slightly, e.g., GPT4's performance rises to 11.7%.
  arXiv  Detail & Related papers  (2024-06-18T00:44:58Z)
• MiniCheck: Efficient Fact-Checking of LLMs on Grounding Documents [62.02920842630234]
  We show how to build small fact-checking models that have GPT-4-level performance but for 400x lower cost. We do this by constructing synthetic training data with GPT-4, which involves creating realistic yet challenging instances of factual errors. For evaluation, we unify datasets from recent work on fact-checking and grounding LLM generations into a new benchmark, LLM-AggreFact.
  arXiv  Detail & Related papers  (2024-04-16T17:59:10Z)
• Evaluating LLMs at Detecting Errors in LLM Responses [30.645694514606507]
  This work introduces ReaLMistake, the first error detection benchmark consisting of objective, realistic, and diverse errors made by LLMs. We use ReaLMistake to evaluate error detectors based on 12 Large Language Models.
  arXiv  Detail & Related papers  (2024-04-04T17:19:47Z)
• TAT-LLM: A Specialized Language Model for Discrete Reasoning over Tabular and Textual Data [73.29220562541204]
  We consider harnessing the amazing power of language models (LLMs) to solve our task. We develop a TAT-LLM language model by fine-tuning LLaMA 2 with the training data generated automatically from existing expert-annotated datasets.
  arXiv  Detail & Related papers  (2024-01-24T04:28:50Z)
• Instances Need More Care: Rewriting Prompts for Instances with LLMs in the Loop Yields Better Zero-Shot Performance [11.595274304409937]
  Large language models (LLMs) have revolutionized zero-shot task performance. Current methods using trigger phrases such as "Let's think step by step" remain limited. This study introduces PRomPTed, an approach that optimize the zero-shot prompts for individual task instances.
  arXiv  Detail & Related papers  (2023-10-03T14:51:34Z)
• Do-Not-Answer: A Dataset for Evaluating Safeguards in LLMs [59.596335292426105]
  This paper collects the first open-source dataset to evaluate safeguards in large language models. We train several BERT-like classifiers to achieve results comparable with GPT-4 on automatic safety evaluation.
  arXiv  Detail & Related papers  (2023-08-25T14:02:12Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.