Task hallucination · AI Developer Accelerator

29d ago in CrewAI

Task hallucination

How do you ensure that a task output does not regurgitate the example you set in the expected_output?

I'm giving an example of a pydantic object in the expected_output of a task, and whenever the task doesn't manage to generate its own object, it tends to make up data, reusing the info from the example I provided in the expected_output. No matter what I say in the tool, task, and agent.

Is there a way to ensure that the agents can more easily distinguish example outputs from expected outputs?

0 comments