Recipes for Thought

The post discusses two approaches to interacting with large language models (LLMs): the 'chat' approach, which involves a traditional back-and-forth conversation, and the 'recipe' approach, which sees each interaction as steps in a structured process. The author suggests that the recipe approach can lead to better results by refining the prompts and shaping the conversation turns more deliberately. This method engages metacognition, helping users understand their own and the LLM's thinking processes. The post also highlights how repeatable thinking recipes can scale cognitive know-how and improve productivity.