1 comment

[ 3.3 ms ] story [ 12.6 ms ] thread
Nice! Overall, frameworks that re-prompt an LLM with feedback or failure modes of the original outputs do very well.