3 ms·

This is a really interesting take, but I don't completely buy it. Traditionally technologies like these become more accessible with time. Take the youtube exa

This is a really interesting take, but I don't completely buy it. Traditionally technologies like these become more accessible with time. Take the youtube example, this is actually a really good counterpoint as if you can create videos with an AI you created, nobody has any way of figuring out that you trained it on so-and-so's IP. It's completely black box which is the direction things are probably going to go IMO. On the other hand, this is problematic because it'll slow down the development of the self-hostable models outside of the community moving to a darkweb like model where they stay beyond the reach of the legal system.

jerf 3y ago
One of the big factors is whether or not AI can stay in the hands of the public or if it recedes into levels of hardware that only companies can own. If the latter, then the control will win. Keeping AI performance workable on personal-level hardware is going to be a big deal. Open source work on keeping performance at that level may be long-term more important that this or that feature. Watch out for calls for AI ethics to turn into calls for controls on hardware capable of running AI. I'm half surprised we aren't already hearing calls for that. I wouldn't be surprised to see it by the end of the year, though my expectation would be at least next year if not the one after. The real goal behind hardware control will be this sort of thing (though copyright control may actually be a secondary concern versus just keeping the really good tools out of the hands of the plebs), not AI ethics. (The elites know AI ethics limitations have no meanings as they have no intention of limiting themselves.)
- gooob 3y ago
  everyone already has computers that can "run AI". would be pretty ridiculous if some law was passed and everyone's computers were confiscated. that would be like a lobotomization of society. a very stupid and short-sited move i think.
- visarga 3y ago
  > One of the big factors is whether or not AI can stay in the hands of the public or if it recedes into levels of hardware that only companies can own. If the latter, then the control will win. Both will win. As corporations will make more and more advanced models, training data generated by their models will uplift open models. They only lag a few months. We already can shrink most of chatGPT's abilities in a 7B model. All open models trained after 2022 are benefiting from their big brothers. It's too easy to exfiltrate skills from proprietary models.

jerf 3y ago

One of the big factors is whether or not AI can stay in the hands of the public or if it recedes into levels of hardware that only companies can own. If the latter, then the control will win. Keeping AI performance workable on personal-level hardware is going to be a big deal. Open source work on keeping performance at that level may be long-term more important that this or that feature. Watch out for calls for AI ethics to turn into calls for controls on hardware capable of running AI. I'm half surprised we aren't already hearing calls for that. I wouldn't be surprised to see it by the end of the year, though my expectation would be at least next year if not the one after. The real goal behind hardware control will be this sort of thing (though copyright control may actually be a secondary concern versus just keeping the really good tools out of the hands of the plebs), not AI ethics. (The elites know AI ethics limitations have no meanings as they have no intention of limiting themselves.)

gooob 3y ago
everyone already has computers that can "run AI". would be pretty ridiculous if some law was passed and everyone's computers were confiscated. that would be like a lobotomization of society. a very stupid and short-sited move i think.
visarga 3y ago
> One of the big factors is whether or not AI can stay in the hands of the public or if it recedes into levels of hardware that only companies can own. If the latter, then the control will win. Both will win. As corporations will make more and more advanced models, training data generated by their models will uplift open models. They only lag a few months. We already can shrink most of chatGPT's abilities in a 7B model. All open models trained after 2022 are benefiting from their big brothers. It's too easy to exfiltrate skills from proprietary models.