Code is easy, writing is hard (for a robot)

Robin Sloan: Fable is very good

Yet even the funkiest JavaScript func­tion car­ries within it many fewer choices than a para­graph of prose. (How’s that for a sentence?) Fable can’t match my writing style yet — honestly, I think that’s beyond the reach of these models, because it’s just too much to simulate, a whole human mind and body, their whole his­tory together.

Fully agreed. I think LLMs are remarkable technology, and they’re incredibly good at writing code. I even think they’re really good at writing boilerplate text for things like documentation. However, while I’ve seen LLMs generate text that takes the form of a Birchtree blog post, they simply can’t get close to the real deal. I’m not saying they write with worse structure or make more mistakes than me (I make way more mistakes than it does), I’m saying it simply doesn’t capture my voice. At its very best, it sounds like a robot pretending to be me, but who’s fooling nobody.