EvoBelief is a self‑evolving world model that represents environment knowledge as falsifiable hypotheses, allowing an LLM agent to update its beliefs during deployment. The system generates hypotheses when unexpected outcomes or unknowns arise, validates them through observations and controlled experiments, and then uses the confirmed rules to guide predictions or plans. Across benchmarks such as ALFWorld, MiniGrid, and DiscoveryWorld, EvoBelief improves task performance, revises stale rules, and discovers quantitative laws, outperforming baseline frozen world models.
By Janet Wang, Jiang Guo, Yi Xiang, Jiaqing Yuan, Mingwen Dong, Hideo Kobayashi, Shuaichen Chang, Chandan K. Reddy, Jiarong Jiang
The paper introduces StepLearn, a nonparametric framework for prequential test‑time learning in large language model agents. StepLearn separates immediate use of informative transitions from persistent trust, turning each transition into a hypothesis that guides the next step and only reusing it after prospective validation across episodes. Experiments on WebArena‑Lite and ALFWorld show StepLearn improves success rates by 2.2–12.7 percentage points over the strongest baseline, with benefits evident from the first task attempts.
By Tong Zhao, Reed Li, Yuyang Hu, Yutao Zhu, Haijin Liang, Haibo Shi, Yu Lu, Zhicheng Dou
The paper introduces Evolving Functional REpertoires (EFRE), a method that replaces a single prompt with a dynamic set of functions to enable large language models to learn new tasks sequentially without forgetting. EFRE refines existing functions when updates are compatible and creates new ones when updates conflict, thereby mitigating catastrophic forgetting and overfitting. Experiments on a three‑task continual‑learning stream show EFRE outperforms GRPO by 7.50 percentage points and reduces performance drop on a subsequent task by only 1.56 percentage points compared to base prompt optimization.
By Fengyuan Liu, Yue Wang, Hangxi Guo, Fengyuan Liu, Chenxu Wu, Yanguang Liu, Mengnan Du
arXiv:2607. 29468v1 Announce Type: new Abstract: Self-play agents can generate training problems without questions from target benchmarks, but their curricula lack persistent state: failures affect gradients yet do not explicitly shape future practice.
By Zenghuang Fu, Zhaoyang Li, Qiuyuan Ai, Haoyu Wu, Minghui Wu, Chenxu Zhao, Ante Wang, Guannan He, Changwei Wang
The paper introduces Update‑Matched Unlearning (UMU), a method that applies the same simulated relearning update to both a defended model and an undefended reference to better preserve plasticity while resisting relearning of deleted knowledge. Experiments on Llama and Qwen models with TOFU and FaithUn datasets show that UMU reduces harmful and benign relearning by 30.4 and 9.9 points on average and increases new‑task accuracy by 3.5 points, supporting the reference‑timing hypothesis. The study demonstrates that robust unlearning should selectively prevent recovery of deleted knowledge without suppressing the model’s ability to learn new tasks.
By Jie Li, Youming Tao, Bowei He, Meng Ding
The paper explores Retrospection-Only Fine-Tuning (ROFT), a method where a language-model agent improves its behavior by generating and training on explanations of its own experiences, without external teachers or reward signals. In software‑engineering tasks with Qwen3.5‑4B, ROFT achieves comparable or better solve rates than GRPO while requiring fewer updates and training time, and can learn from failures alone. Behavioral analysis shows ROFT indirectly assigns credit to actions and can produce shorter, more direct solutions when prompted to focus on direct solutions.
arXiv:2610.13991v1 Announce Type: cross
Abstract: Self-evolving agents improve through repeated cycles of modification and evaluation. However, existing methods receive feedback on modifications only...
By Zixuan Xie, Chang Xu, Yuhua Jiang, Li Zhao
The paper introduces Boundary-Aware Skill Memory (BASM), a method that enriches skill memories for large language model agents with explicit boundary fields such as applicability conditions, risk cues, avoidance rules, and recovery notes. This approach transforms retrieved skills from unconditional templates into state‑conditioned guidance, preventing the Skill Imitation Trap where more skills lead to incorrect tool usage. Experiments on three agent benchmarks and four model scales show that BASM improves task success rates, accuracy, and reduces attack success while cutting average steps compared to memory‑free baselines.
By Zihan Lin, Zhenyu Chen, Jiawen Wei, Xiaohan Wang, Jie Cao, Jiajun Chai, Wei Lin, Guojun Yin, Ran He
arXiv:2607. 14408v1 Announce Type: new Abstract: A self-evolving agentic loop repeatedly proposes a tweaked version of an agent (its prompt template or program) and accepts or rejects the change based on a per-iteration quality signal.
By Minghao Liu, Yu Wang, Jiayun Wang, Wei Wei
FREEEVOLVE automates the optimization loop used by agent evolvers, allowing the system to decide what to test, how much evidence to gather, which candidates to pursue, and when to stop, all within specified resource limits. By treating these decisions as an editable evolution skill and improving them through meta‑evolution, the method learns a capability that replaces hand‑crafted search loops. Experiments on several benchmarks show that FREEEVOLVE improves primary metrics by an average of 13.6 points and that meta‑evolved skills add further gains, indicating transferability across environments.
By Lecheng Kong, Like Hui, Nikos Kanakaris, Prithwish Jana, Sahika Genc, Narayanan Sadagopan
arXiv:2607. 01480v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards (RLVR), along with recent selfdistillation variants such as SDPO, evaluates each rollout against a verifier and updates the policy from that episode-level signal.
By Ye Liu, Srijan Bansal, Bo Pang, Yang Li, Zeyu Leo Liu, Yifei Ming, Zixuan Ke, Shafiq Joty, Semih Yavuz
SkillNEO introduces a non‑elitist evolutionary framework for refining agent skills, allowing lower‑reward versions to receive further edits and preserve useful partial repairs. By separating the selection of a version to use from the selection of versions to improve, the method identifies which steps to keep and which failures to fix, even revisiting earlier versions with new evidence. Experiments on 86 SkillsBench tasks across eight domains show significant reward gains for GPT‑5.4 Mini and substantial improvements when the evolved skills are reused by stronger models such as GPT‑5.5 and GPT‑5.6 Luna.
By Enze Ma, Chunyu Miao, Yankai Chen, Jie Yang, Liancheng Fang, Kening Zheng, Hanrong Zhang, Muyao Xu, Xinle Yu, Chengze Li, Wei-Chieh Huang, Ziwei Ren, Yunhao Du, Zhixiang Shen, Xue Liu, Philip S. Yu, Zhen Wang