Post #1865898

2026-02-17 03:18 UTC

Highly-trained LLMs are a form of capital. They are produced by extracting knowledge from text which workers have made, and they cannot be made practically useful without vast compute expenditures. The companies which make them seek to recoup their costs by renting them out, primarily to corporations which want to use them to reduce the number of workers which they employ. The facts are straightforward, and the net effect is a wealth transfer from workers to owners. Current attempts at producing large-scale social or economic transformations via generative AI are an unavoidably right-wing political project.

Replies (3)

@jnkrtech@social.treehouse.systems 2026-02-17 03:32

Unlike cryptocurrency-based smart contracts, I don’t think that machine learning and generative AI are inherently right-wing technologies. It’s totally possible to encode your own little crappy neural net and have it spit out weird pixels for an art project, and that’s okay. I honestly don’t think that LLMs are a terrible tool for search and data extraction, and I’ve seen reasonable claims for their usefulness in other circumstances as well. The key point is that a small list of targeted use-cases is not going to upend society. Generative AI is being marketed as a revolutionary force which is going to somehow destabilize and remake our whole economic system. This is because the most capable and expensive generative AI systems are produced and controlled by capitalists who have destabilization and wealth capture as their goal. Any AI tools which are capable of causing significant economic impacts have been produced to further this objective.

Open ##1865899
@Viss@mastodon.social 2026-02-17 13:05

@jnkrtech i have a shitload of research im doing on this topic you may appreciate. theres a bunch od math involved and a bunch of live fire tests i need to do

Open ##1865903
@johntinker@hear-me.social 2026-02-17 17:32

@jnkrtech The LLM is a statistical analysis of a corpus of language. In a sense, it is language in, language out, with math in the middle. The math is done on the corpus. The "front end" is parsing user intent against the business logic that built the machine, and inducing output from the LLM. For those who are interested in language itself, and how language has been used, it is a tool of considerable impact, I think.

Open ##1865904