A new arXiv paper introduces SimSkill, a self-evolving agent built around the SUMO traffic simulator. The system targets lifelong learning: turning accumulated experience into durable, reusable competence rather than solving one-off requests. It matters to practitioners exploring LLM agents that compound skills over time in simulation environments.
Researchers introduced ContinualSkillBench, a dynamic evaluation framework for in-context continual skill learning in LLM agents. It covers five domains with 100 interconnected subtasks each, ordered by difficulty and cross-task reuse. Early experiments suggest sequential execution struggles, highlighting gaps in current agent skill evolution.
Researchers introduced ContinualSkillBench, a dynamic evaluation framework for in-context continual skill learning in LLM agents. It covers five domains with 100 interconnected subtasks each, ordered by difficulty and cross-task reuse. Early experiments suggest sequential execution struggles, highlighting gaps in current agent skill evolution.
A new paper introduces TRACE, a self-evolving skill bank that helps LLM agents maintain consistency and recognize their limits, addressing a reliability gap exposed by CAR-bench, a benchmark for in-car assistants. The approach enables agents to handle ambiguous user requests through multi-turn dialogue and tool use while adhering to domain policies.
A new paper introduces TRACE, a self-evolving skill bank that helps LLM agents maintain consistency and recognize their limits, addressing a reliability gap exposed by CAR-bench, a benchmark for in-car assistants. The approach enables agents to handle ambiguous user requests through multi-turn dialogue and tool use while adhering to domain policies.
A new arxiv paper proposes Skill Self-Play, a method where LLMs co-evolve skills through interaction, reconciling the tension between task diversity and verification reliability. The approach uses agent skills as a middle ground, enabling deep, verifiable execution in specific domains while broadening the task space. This could advance self-evolutionary training for LLMs.
A new arxiv paper proposes Skill Self-Play, a method where LLMs co-evolve skills through interaction, reconciling the tension between task diversity and verification reliability. The approach uses agent skills as a middle ground, enabling deep, verifiable execution in specific domains while broadening the task space. This could advance self-evolutionary training for LLMs.
A new arXiv paper argues that among skill-learning strategies for adapting LLM agents to novel domains, treating skills as programs yields the greatest cost reduction. By executing action sequences deterministically, program-augmented agents can reliably and cheaply achieve goals that would otherwise be expensive. The work highlights cost effectiveness over raw performance, a shift from prior skill-learning research.
A new arXiv paper argues that among skill-learning strategies for adapting LLM agents to novel domains, treating skills as programs yields the greatest cost reduction. By executing action sequences deterministically, program-augmented agents can reliably and cheaply achieve goals that would otherwise be expensive. The work highlights cost effectiveness over raw performance, a shift from prior skill-learning research.
An arXiv paper, PRACTICE: From Experience to Expertise in Self-Evolving Embodied Agents, addresses how multimodal LLM-based embodied agents can continually improve through interaction. It argues existing experience-based methods rely on manually designed prompting workflows to extract and update skills, which struggle to adapt to new and diverse environments. The work matters to practitioners building agents that must adapt rapidly rather than follow fixed skill-extraction procedures.
An arXiv paper, PRACTICE: From Experience to Expertise in Self-Evolving Embodied Agents, addresses how multimodal LLM-based embodied agents can continually improve through interaction. It argues existing experience-based methods rely on manually designed prompting workflows to extract and update skills, which struggle to adapt to new and diverse environments. The work matters to practitioners building agents that must adapt rapidly rather than follow fixed skill-extraction procedures.