I like the author's idea that people should publish the prompts they use to generate LLM output, not the output itself.