Large language models (LLMs) can produce erroneous responses that sound fluent and convincing, raising the risk that users will rely on these responses as if they were correct. Mitigating such overreliance is a key challenge. Through a think-aloud study in which participants use an LLM-infused application to answer objective questions, we identify several features of LLM responses that shape users' reliance: explanations (supporting details for answers), inconsistencies in explanations, and sources. Through a large-scale, pre-registered, controlled experiment (N=308), we isolate and study the effects of these features on users' reliance, accuracy, and other measures. We find that the presence of explanations increases reliance on both correct and incorrect responses. However, we observe less reliance on incorrect responses when sources are provided or when explanations exhibit inconsistencies. We discuss the implications of these findings for fostering appropriate reliance on LLMs.

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.

The Internet has enabled learning at scale, from Massive Open Online Courses (MOOCs) to Wikipedia. But online learners may become passive, instead of actively constructing knowledge and revising their beliefs in light of new facts. Instructors cannot directly diagnose thousands of learners' misconceptions and provide remedial tutoring. This paper investigates how instructors can prompt learners to reflect on facts that are anomalies with respect to their existing misconceptions, and how to choose these anomalies and prompts to guide learners to revise incorrect beliefs without any feedback. We conducted two randomized experiments with online crowd workers learning statistics. Results show that prompts to explain why these anomalies are true drive revision towards correct beliefs. But prompts to simply articulate thoughts about anomalies have no effect on learning. Furthermore, we find that explaining multiple anomalies is more effective than explaining only one, but the anomalies should rule out multiple misconceptions simultaneously.