Mind Your Step (by Step): Chain-of-Thought can Reduce Performance on Tasks where Thinking Makes Humans Worse

Chain-of-thought (CoT) prompting has become a widely used strategy for improving large language and multimodal model performance. However, it is still an open question under which settings CoT systematically reduces performance. In this paper, we seek to identify the characteristics of tasks where CoT reduces performance by drawing inspiration from cognitive psychology, focusing on six representative tasks from the psychological literature where deliberation hurts performance in humans. In three of these tasks, state-of-the-art models exhibit significant performance drop-offs with CoT (up to 36.3\% absolute accuracy for OpenAI o1-preview compared to GPT-4o), while in others, CoT effects are mixed, with positive, neutral, and negative changes. While models and humans do not exhibit perfectly parallel cognitive processes, considering cases where thinking has negative consequences for humans helps identify settings where it negatively impacts models. By connecting the literature on human verbal thinking and deliberation with evaluations of CoT, we offer a perspective for understanding the impact of inference-time reasoning.

We find that chain-of-thought, a method for allowing LLMs to generate reasoning before providing responses, can lead to worse outcomes in similar failure cases to when people overthink. We do this by taking a broad literature in psychology on human overthinking and applying these tests to various AI models, upon which we find some drastic decreases in model performance. Our paper shows that allowing an AI to reason before responding is not always beneficial; instead, it is important to look at the specific use case and maybe even take a jab at it yourself.