LLMs seem to have worse intuition than experts and compensate by being able to cover a much wider surface area of ideas, so we might just need to extract the intermediate progress along the way.