Can One Model Run Your App And Evals?
Hamel Husain · 3:16
Using the same model as both your agent and your LLM judge is usually fine. The judge does a different, narrower job, and what matters is whether its verdicts match human judgments on held-out data. Only switch models...