Can Large Language Models Always Solve Easy Problems if They Can Solve Harder Ones?
This addresses a critical reliability issue for users of LLMs in applications requiring consistent reasoning, though it is incremental as it builds on existing inconsistency research.
The paper tackles the inconsistency problem in large language models (LLMs), where they can solve hard problems but fail at easier ones, by developing the ConsisEval benchmark and introducing a consistency score, finding that GPT-4 achieves a 92.2% consistency score but still has specific failures.
Large language models (LLMs) have demonstrated impressive capabilities, but still suffer from inconsistency issues (e.g. LLMs can react differently to disturbances like rephrasing or inconsequential order change). In addition to these inconsistencies, we also observe that LLMs, while capable of solving hard problems, can paradoxically fail at easier ones. To evaluate this hard-to-easy inconsistency, we develop the ConsisEval benchmark, where each entry comprises a pair of questions with a strict order of difficulty. Furthermore, we introduce the concept of consistency score to quantitatively measure this inconsistency and analyze the potential for improvement in consistency by relative consistency score. Based on comprehensive experiments across a variety of existing models, we find: (1) GPT-4 achieves the highest consistency score of 92.2\% but is still inconsistent to specific questions due to distraction by redundant information, misinterpretation of questions, etc.; (2) models with stronger capabilities typically exhibit higher consistency, but exceptions also exist; (3) hard data enhances consistency for both fine-tuning and in-context learning. Our data and code will be publicly available on GitHub.