After watching Karpathy’s new intro video to LLMs it’s apparent why prompt engineering isn’t a thing: you’re just training intuition in a System 1 mind (what we’d call Preconscious in story). The next step, likely all this Q* stuff, will be an API for a System 2 mind (Conscious) where–regardless of prompt–the LLM will do off and think through the question/problem step by step and then work out the best answer using test time compute. 🤓