#LLM Judge

Netflix says subjective AI evals need two judges

Netflix's Cameron Wolfe tells AI Engineer that subjective AI tasks drown in noise and need human evals, tuned LM judges, and production AB tests.