[keiferski]

Two thoughts:

1. The default LLM behavior (at least what I’ve used as a consumer with ChatGPT Claude etc.) is to be excessively verbose, presumably because costs are tied to usage and therefore the assumption is that more text = better.

I’ve spent over a decade working as a copywriter, and IME the most important part of writing is the edit - what to cut out.

So I think it’s probably possible that an AI could write stuff we’d want to read, the default behavior of the AIs most people are using works against it.

2. A lot of the writing that actually gets read today is either a description of a lived experience and/or involves slang. Neither of those things are interesting if done by an AI - I don’t care about the imagined experience of an LLM.