In Re: A comment on a discord about how AI coding assistants maintain thir userbase by staying marginally ahead of the curve and selling compute, both unsustainable options for longterm profitability, and thus such things are destined to become like grammarly to spellcheck.

The only potential issue with that is role that specialized datasets could play in future performance, just training something on github is fine for writing code but they need data on how AI/users deploy programs (think dependancy mgmt), since that is the more fiddly stuff that ime AI tends to really struggle with. If it turns out that 1. current AI companies are building those datasets and 2. such datasets are actually useful then they could have a monopolized resource necessary for coding AI production

Certainly beyond what the og article was claiming but I think it would be deeply unwise to just be like "the tech will sort itself" since the tech will seek refuge (monopoly power) in extant structures desinged to create it (notions of proprietary user data)

no usefull tech under capitalism and such