A new test finds that AI agents do not work each time with emotional users
Data from the Tracer pipeline improves Qwen3 models at each size of model in the test.
Researchers made Tau-Emotion-Bench, a test for agents that use tools, in 10 domains with users who show different emotions. The test used 2 user simulators. The agents in the test have a large gap between average success on tasks and success on each try.
Sources
Posted
