智能体与自主科学
突破级
暂无讲解视频
收录解读
Continual Harness: Online Adaptation for Self-Improving Foundation Agents 关注的是一个可复用的 AI 系统或评测问题,而不是单点 demo。
Online harness refinement for self-improving foundation agents.
It formalizes the agent harness as an adaptive object with trajectory-driven refiners, extending manual harness repair into an online self-improvement loop.
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。