The idea that the text outputs of LLMs--built on text of course--is "chat," speech, is part of how the idea that we constantly need to circumvent existing text by generating new disposable text gets naturalized. Nothing wrong with speech, of course, but the point of text is that it can persist.

It feels like a very TechBro mindset that "nothing that already exists has value - we have to disrupt, remix, re-invent the wheel from first principles"

"Chat" is critical in adoption of LLMs in tech, where they're as an existential threat to jobs as the arts. The use to date is not so much in code generation but as a replacement for stack overflow. The ability to ask for questions, clarifications and examples after the initial test is a key feature

@ncecire.bsky.social it’s a kind of inversion of “do your research”: “please provide me an authoritative-sounding summary so I don’t reflect on its origin in webpages by randos”