智能体与自主科学
突破级
暂无讲解视频
收录解读
Do Self-Evolving Agents Forget? Capability Degradation and Preservation in Lifelong LLM Agent Adaptation 关注的是一个可复用的 AI 系统或评测问题，而不是单点 demo。
Capability erosion and preservation in lifelong self-evolving LLM agents.
It identifies a core failure mode of self-evolving agents and proposes capability-preserving evolution across workflow, skill, model, and memory channels.
它没有更高，是因为这些新 arXiv 工作仍需要更多独立复现、真实系统部署和长期社区采用来确认影响。