255
Meta will train AI with data from European users
(stackdiary.com)
This is a most excellent place for technology news and articles.
There's already some open source models from other companies. I also think the requirements will go down with time as well but energy is definitely an issue.
Meta actually released an open source model which jumpstarted the whole ecosystem, anyone can fine-tune a base model now. You can take your favorite hobby, accumulate data on it and build something in a few days and share it.
I just think the good outweigh the bad and individuals weren't going to get paid anyways. Most of the data is owned by specific websites, big publishing houses and the like so I can overlook the infringement issues.
I don't know man. I mean, if you assume that the people that were doing the art, music, code, etc, that's being stolen were not going to get paid anyway then yeah. If they were doing shit just for love they may continue, and with new toys lol.
But I don't think that's a good assumption. Even if not a lot, sone people do get paid for this kind of work. And now they will not get paid anymore. And maybe that's leveling the playing field. Or maybe that's telling people with talent to stop doing what they do well. Probably both. But at the end of the day we are going to see less art made by people and more done by "AI", much more done by "AI".
And that's the biggest problem IMO, for most people art is social and part of the reward is the recognition we get from other people when we do good art. But with AI that's gone, on the internet at least. The sea or superficially good but mediocre shit we are already seeing is going to kill a lot of indie art.
And then there is hallucinations which seem unsolvable ... and the environmental damage ... and the labor practices abuses ... and their monopolization of the technology ... and their missleading marketing ... I honestly just see so much damage and almost no benefit, yet ... maybe some day it all will pay off, I don't see it