Benchmarking Reference-Free LLM Agent Robustness Under Schema, Policy, and Toolset Drift

Tool-using large language model (LLM) agents fail when external interfaces evolve, yet most evaluations emphasize static competence rather than adaptation under drift. We studied this problem with controlled schema, policy, and toolset perturbations derived from tau2-bench retail tasks and distingui...

Celý popis

Podrobná bibliografie

Hlavní autoři:	Mohammad Hasbi Assidiqi, Daniyal Alghazzawi, Suaad Alarifi, Li Cheng
Médium:	Artigo
Jazyk:	Inglês
Vydáno:	IEEE 2026-01-01
Edice:	IEEE Access
Témata:	Continual learning drift adaptation large language models benchmarking tool use
On-line přístup:	https://ieeexplore.ieee.org/document/11534189/
Tagy:	Přidat tag Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!