Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRs

By K. Gandhi, Ayush K Chakravarthy, Anikait Singh, nathan lile, Noah D. Goodman. arXiv.org, 2025

413 citations recorded by the source corpus.

Test-time inference has emerged as a powerful paradigm for enabling language models to ``think''longer and more carefully about complex challenges, much like skilled human experts. While reinforcement learning (RL) can drive self-improvement in language models on verifiable tasks, some models exhibit substantial gains while others quickly plateau. For instance, we find that Qwen-2.5-3B far exceeds Llama-3.2-3B under identical RL training for the game of Countdown. This discrepancy raises a critical question: what intrinsic properties enable effective self-improvement? We introduce a framework to investigate this question by analyzing four key cognitive behaviors -- verification, backtracking, subgoal setting, and backward chaining -- that both expert human problem solvers and successful language models employ. Our study reveals that Qwen naturally exhibits these reasoning behaviors, whereas Llama initially lacks them. In systematic experimentation with controlled behavioral datasets, we find that priming Llama with examples containing these reasoning behaviors enables substantial improvements during RL, matching or exceeding Qwen's performance. Importantly, the presence of reasoning behaviors, rather than correctness of answers, proves to be the critical factor -- models primed with incorrect solutions containing proper reasoning patterns achieve comparable performance to those trained on correct solutions. Finally, leveraging continued pretraining with OpenWebMath data, filtered to amplify reasoning behaviors, enables the Llama model to match Qwen's self-improvement trajectory. Our findings establish a fundamental relationship between initial reasoning behaviors and the capacity for improvement, explaining why some language models effectively utilize additional computation while others plateau.

View Paper: https://www.semanticscholar.org/paper/432cf4506d1fa7773abfaceb35841d60e5685839

Citations in the corpus

• Emerging Properties in Unified Multimodal Pretraining (2025, unknown): “5 LLM [93] as the initialization for its superior performance [21] and public availability.”
• Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning (2025, unknown): “Our finding that RLVR primarily focuses on forking tokens in reasoning paths may share some common ground with the observations of Gandhi et al. (2025) and Li et al. (2025), who suggest that RLVR primarily learns the format rather than the content.”
• SimpleRL-Zoo: Investigating and Taming Zero Reinforcement Learning for Open Base Models in the Wild (2025, unknown): “We apply Gandhi et al. (2025)’s cognitive behavior framework to perform a detailed analysis of how model reasoning behaviors change during ”zero training.””