Build the evaluator-optimizer loop, where one LLM call generates and another critiques — when to use it, the judge's biases, the stop condition people get wrong, its research roots, and when not to loop.
Loading content...