I don’t think this is true, the models will become much better at prompting themselves, I already do this with GPT-4, I would assume the same works for any LLM.

→ View original post on X — @officiallogank