My best guess is that the argument is “the scale of data needed to train an LLM can never be obtained in a manner that isn’t oppressive or stealing by it’s scale” and thats not really true. there are a lot of open works, works which allow using them in datasets, etc. there are ways to get permission from people, give attribution, and just be honest about everything. the reason its not happening is because we live in a hell of a world with oppressive systems that benefit from hiding and manipulation

there’s no reason it cant be better in a better world

if the argument is generating things like drawings, renders, writing, music, videos, are inherently bad than that’s not really saying all LLMs are bad

accessibility LLMs should be a part of a better world at least. what changed my perspective is seeing someone irl who didnt speak english and just voice messaged chatgpt in their language and it translated that to me in english. was it perfect? i doubt that. do i support chatgpt? fuck no. do i believe something like this can be done properly and for public good? yes