CORTEXA
← Browse

Yuze Shi

2 papers indexed

arxivcs.LGcs.AI2026-07-03

No Time Like the Present: Agentic Test-Time Training for LLM Agents

Yanbo Wang, Jinhua Hao, Yuze Shi, Kun Yuan, Ming Sun

LLM agents often degrade over long episodes: as trajectories grow, they revisit explored states, repeat failed actions, and lose strategies that previously worked. Test-time training (TTT) offers a way to adapt model weights to the evolving task state, but existing LLM TTT method…

View free PDFSource page