From how datasets contain a good amount of pirated stuff
Personally, I do not care if datasets contain pirated stuff because the copyright laws are broken anyway. If the entirety of Disney movies and Harry Potter books are somewhere inside those datasets, I can play them a song on the world's smallest violin.
Smaller artists/writers are the ones I empathize with. I get their concern about large corporations using their stuff and making money off of it. Not entirely something that applies to local AI since most people here do this for themselves and do not make any money out of it.
to the environmental impact
That's actually the saddest part. Those models could be easily trained with renewables alone but you know, capitalism.
Do we do the same unhealthy stuff with it, or do we help even out the playing field so it’s not just the mega-corporations in control of AI?
The thing is, those models are already out there and the people training them do not gain anything when people download their open weights/open source models for free for local use.
There's so much cool stuff you can do with generative AI fully locally that I appreciate that they are available for everyone.
Second controversial take: I think AI isn’t very intelligent.
If we are talking about LLMs here, I don't think that's much of a controversial take.
Most people here will be aware that generative AI hallucinates all the time. Sometimes that's good, like when writing stories or generating abstract images but when you're trying to get accurate information, it's really bad.
LLMs become much more useful when they do not have to completely rely on their training data and instead get all the information they need provided to them (e.g. RAG).
I'm a huge fan of RAG because it cites where it got the information from, meaning you can ask it a question and then continue reading in the source to confirm. Like fuzzy search but you don't have to know the right terms.