I spent the last month writing with ChatGPT, and found some better ways to steer it.

After spending a lot of time on it, I believe ChatGPT’s generic writing is actually just a steering problem.

I believe you need A LOT more context than people expect. ~5k tokens for a 1k token blog post.

The big question is "How do you give an LLM enough context to steer it to write what you wanted to?"

Using voice to capture LOTS of raw tokens / notes, then refining them through AI is working pretty well for me, and I'd love to see what other people think.

There’s an interactive demo to try it out. I’d love your feedback and opinions!