The best reason to not use AI in writing is similar to the reason why lecturing with a whiteboard is better than slides. Humans absorb complex information at about the rate humans can write it on a whiteboard. But if you make good slides accounting for this factor, and modulate speech tempo, the difference vanishes and you get to use more types of info like pictures or diagrams that would take too long to draw live.

Bad slides create rate overload and inject noise. What gets through is random samples of what you put out.

Engineering transmission rate limits and noise suppression into AI generated text is a similar challenge. I’m slowly getting better at it.

Back when I was teaching regularly I used to be a bit of a whiteboard fundamentalist (chalkboard even) and avoided slides or transparencies. Then I learned rate and noise disciplines. Methods like Lessig and pecha-kucha helped. Now I just freestyle it and even do calibrated overload slides for certain uses.

Learning good AI writing skills feels very similar to learning good slide presentation skills. There are more ways things can go wrong but also more ways things can go right.