You are building a customer-facing assistant that answers account and policy questions from a language model. The team has seen cases where the model gives confident but unsupported responses, which creates risk in a support setting.
How do you mitigate hallucination in a generative model used for customer-facing applications?
You are building a customer-facing assistant that answers account and policy questions from a language model. The team has seen cases where the model gives confident but unsupported responses, which creates risk in a support setting.
How do you mitigate hallucination in a generative model used for customer-facing applications?
