TAG

#agent reliability

Stories connected through a shared idea, actor, technology or theme.

16 AugCAR-bench tests whether agents recognise their limitsThe benchmark tests multi-turn task completion, hallucination resistance when capabilities are missing and disambiguation of ambiguous requests. Agents must chain tools and policies while interacting with a simulated user in a changing environment.→