Post

I've found success in complex LLM projects by separating all of parts and having it work on multiple roles discretely. Prompt first for an approach. Then prompt for an outline of work. Then prompt for each section in the outline. Etc.