Have you experimented with weighting the self-evaluations based on specific criteria (e.g., correctness, clarity, creativity), or using external validators to g

Have you experimented with weighting the self-evaluations based on specific criteria (e.g., correctness, clarity, creativity), or using external validators to guide the AI’s final choice? Curious how much tuning the evaluation step impacts overall performance.