Post

LLMs are likely going to be the mediating lens through which large swathes of the population make sense of the world around them on political issues, & it's very bad that a man who has a strong case for being one of the worst people alive is one of the few to be actively trying to exploit this

工an Monroe 🚰

@eean.dev

Another reason to stick with Chinese LLMs Seriously there are few network effects and so far not beating "commodity technology" allegations

28/4/26, 11:07 pm0 reposts32 likes

MVB

@mvb.kirinok.osaka

29 Apr

Those are trained by copying American LLMs atm, so their answers aren't very different

Tanner B 🆓🇵🇸

@nspants.bsky.social

29 Apr

Not in my experience. They're way less biased towards western hegemony imo

Stephen Judkins

@stephenjudkins.bsky.social

29 Apr

yeah American LLMs were trained by copying a bunch of shit they don't necessarily have IP rights for, either. so what?

Alvaro Ibañez

@alvaroim.bsky.social

29 Apr

What are you talking about? The objection is that they are trained on the same BS and, thus, will spew out the same BS as the US ones

工an Monroe 🚰

@eean.dev

29 Apr

It's true that the entire corpus of human written culture includes a lot of BS but I dunno about calling it BS 😅 Model theft is the idea of training a model with another model and this is cheating for Reasons

工an Monroe 🚰

@eean.dev

29 Apr

Google's legal theory of model weight theft is like sovereign citizen-level legal bs imo lol

Stephen Judkins

@stephenjudkins.bsky.social

29 Apr

Yeah! I am very sympathetic to LLM training being IP theft, but it is impossible to square that initial training from a human-generated corpus is not theft, but one LLM training from another is somehow theft