BouncerBench
收藏IEEE2026-04-17 收录
官方服务：
资源简介：
BouncerBench is a benchmark for evaluating the ability of Large Language Models (LLMs) to abstain when presented with ambiguous tasks or when a sufficiently accurate response cannot be provided. Trust and reliability are critical to the success of AI agents in software engineering tasks, and BouncerBench introduces a notion of \bouncers\ that could be applied to any autonomous ticket resolution system.
