The Self-Refinement methodology mimic humans in a three step process where initial outputs from an LLM is refined via a sequence of iterative feedback.