Surprising to see so little traction on this; I hope this makes it to the second chance pool because it would interesting to hear the LLM advocates' take on it.

Is the poor performance because the LLMs are not being used for iterative refinement?