Five prompting habits that Anthropic and OpenAI both put in their own documentation
Both labs publish the same core advice: write the instruction as a command rather than a question, give the model explicit permission to say it does not know, put the context before the question, keep one task per prompt, and tell the model how proactive to be. The common thread is that every one of them removes ambiguity — none of them is a phrase you paste in to unlock hidden quality.
約4分