In this work, we reveal that Large Language Models (LLMs) possess intrinsic behavioral plasticity-akin to chameleons adapting their coloration to environmental cues-that can be exposed through token-conditional generation and stabilized via reinforcement learning. Specifically, by conditioning generation on carefully selected token prefixes sampled from responses exhibiting desired behaviors, LLMs seamlessly adapt their behavioral modes at inference time (e.g., switching from step-by-step reasoning to direct answering) without retraining. Based on this insight, we propose Token-Conditioned Reinforcement Learning (ToCoRL), a principled framework that leverages RL to internalize this chameleon-like plasticity, transforming transient inference-time adaptations into stable and learnable behavioral patterns. ToCoRL guides exploration with token-conditional generation and keep enhancing exploitation, enabling emergence of appropriate behaviors. Extensive experiments show that ToCoRL enables precise behavioral control without capability degradation. Notably, we show that large reasoning models, while performing strongly on complex mathematics, can be effectively adapted to excel at factual question answering, which was a capability previously hindered by their step-by-step reasoning patterns.
When RL Fails after SFT: Rejuvenating Model Plasticity for Robust SFT-to-RL Handoff
Hong Kong University of Science and Technology(香港科技大学);Zhejiang University(浙江大学);State Key Laboratory of Multimodal Artificial Intelligence Systems, CASIA(多模态人工智能系统国家重点实验室,CASIA)
Neural Networks Remember More: The Power of Parameter Isolation and Combination
Aberdeen Institute of Data Science and Artificial Intelligence(阿伯丁数据科学与人工智能学院);South China Normal University(华南师范大学);School of Artificial Intelligence(人工智能学院)
Rethinking Prospect Theory for LLMs: Revealing the Instability of Decision-Making under Epistemic Uncertainty
Hong Kong University of Science and Technology(香港科技大学);Huazhong University of Science and Technology(华中科技大学);University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校)
Beyond BFI: The CSI for Enhanced Reliability and Validity in Evaluating LLM Personality Traits
University of Illinois Chicago(伊利诺伊大学芝加哥分校);Institute of Automation, Chinese Academy of Sciences(中国科学院自动化研究所);Microsoft Research Asia(微软亚洲研究院);North Carolina State University(北卡罗来纳州立大学)
Revealing emergent human-like conceptual representations from language prediction
College of Computer Science and Artificial Intelligence(计算机科学与人工智能学院);Fudan University(复旦大学);Institute of Modern Languages and Linguistics(现代语言与语言学研究所);Shanghai Key Laboratory of Intelligent Information Processing(上海智能信息处理重点实验室);Research Institute of Intelligent Complex Systems(智能复杂系统研究所);Institute of Trustworthy Embodied Artificial Intelligence(可信具身人工智能研究所);State Key Laboratory of Genetics and Development of Complex Phenotypes(复杂表型遗传与发育国家重点实验室)
Deep RL Needs Deep Behavior Analysis: Exploring Implicit Planning by Model-Free Agents in Open-Ended Environments
Department of Neurobiology, Harvard Medical School(哈佛医学院神经生物学系);Kempner Institute for the Study of Natural and Artificial Intelligence, Harvard University(哈佛大学自然与人工智能研究学院);Department of Mathematics, NTNU(NTNU数学系);School of Computer Science, McGill University & Mila(麦吉尔大学计算机科学学院及Mila);Department of Computer Science, University of Toronto(多伦多大学计算机科学系);Biophysics Graduate Program, Harvard University(哈佛大学生物物理学研究生项目)
Better Adherence, Richer Context: A Field Evaluation of LLM-Powered Conversational Voice Diaries for Sleep
The Johns Hopkins University(约翰霍普金斯大学);Department of Psychiatry and Behavioral Sciences, The Johns Hopkins University School of Medicine(精神病学与行为科学系,约翰霍普金斯大学医学院)
From Behavioral Performance to Internal Competence: Interpreting Vision-Language Models with VLM-Lens
University of Waterloo(滑铁卢大学);Vector Institute(向量研究所);University of Michigan(密歇根大学);Stony Brook University(石溪大学);McGill University(麦吉尔大学);Dartmouth College(达特茅斯学院)