I almost always get better results from LLMs by going back and editing my prompt and starting again, rather than trying to correct/guide it interactively. Almos

I almost always get better results from LLMs by going back and editing my prompt and starting again, rather than trying to correct/guide it interactively. Almost as if having mistakes in your context window is an instruction to generate more mistakes! (I'm sure it's not quite that simple)