LATIDIA · Investigación
Harnesses que se mejoran a sí mismos: un marco unificador para los harnesses de agentes que se mejoran a sí mismos
arXiv:2611.22299v1 Tipo de anuncio: nuevo Resumen: Que un modelo fundacional tenga éxito como agente autónomo depende tanto de su «software» de tiempo de ejecución como de sus pesos. Este tiempo de ejecución, el «arnés» del agente, gestiona el contexto y m
WhatsApp ↗Telegram ↗
La noticia
arXiv:2611.22299v1 Announce Type: new Abstract: Whether a foundation model succeeds as an autonomous agent depends as much on its runtime software as on its weights. This runtime, the agent harness, manages context and memory, steers multi-turn control, executes tool actions in sandboxes, and verifies work before stopping. Because models differ in how they use tools and how they fail, a static harness tuned for one model or task often stalls, wastes tokens, or regresses on another. This bottleneck has driven a shift from prompt tuning to self-improving agent harnesses: systems that inspect their own execution traces, separate runtime bugs from model limits, and rewrite their own code. We present a