----
Last week I came down with a fever and decided to run some experiments on whether the new GPT could generate long-form fiction. Around the same time, I saw a HN comment about how technology gains have lead SWEs to work at higher levels of abstraction—and I thought “huh, why not writers?”
Anyway so I was locked indoors for a week and a half and I spent it coding an answer that question.
Most of the magic sauce here is just in making it feel really snappy and good to edit—making it easy for the human to collaborate with the objectively faster LLM. The underlying prompt engineering is cool, but it’s the feeling that you can do more than just shoot a prompt at ChatGPT and hope for a good result, or start writing and hope for a good autocomplete.
Definitely gave me some ideas for future LLM experiences I want to try out, let me know what you think!
