4 ms·

What makes you think this would strongly harm open-source AI projects? Do you mean e.g. small "companies trying to make money?" If so, I see that danger, but I

by jrm4 3y ago

What makes you think this would strongly harm open-source AI projects? Do you mean e.g. small "companies trying to make money?" If so, I see that danger, but I also kind of don't care. What I don't think it would harm is actual "free/free-ish open source, just OUT there" AI projects in terms of pure "software that's available for everyone to mess around with, regardless of legality," -- and (as a lawyer I can't recommend doing this at all of course, but) I think it would end up mostly being a good and safer thing (in the same way that having e.g. Kali Linux out there is a good thing). The models are out there and it's hard to think that any single player could have a real big advantage here. I say let the suits commence.

ronsor 3y ago
It's pretty hard to develop an open-source project in the open if your activity is seen as illegal. We don't need AI projects lumped in with Popcorn Time and movie torrents.
- jrm4 3y ago
  Interesting examples you put there; FWIW you named the things that typically strongly favor the rights and desires of the consumer, despite the law, and give them a significantly superior experience to their legal alternatives (at least in the foreseeable short term.)
stale2002 3y ago
> Do you mean e.g. small "companies trying to make money?" No they mean open source projects. The overwhelming majority of all generative AI related open source projects depend on, use, or are related to publicly available weights that were trained without permission on other people's data. The entire open source AI community wouldn't exist if those weights weren't available for everyone to use for basically any reason. One could argue that the existing weights aren't going to disappear as thats impossible to enforce. But what about the updated weights? If nobody is allowed to spend 50 million dollars training a new model and then releasing that model publicly for anyone to use, without getting permission from the entire internet to make that model, well then there aren't going to be anymore new models coming out. The government isn't going to be able to confiscate everyone's gamer PCs in order to ban inference, but they can stop someone from spending that 50 mil training the next cutting edge open source model. Which means that the existing models will be all that the open source community has. Yeah they could fine tune them, but cutting edge base models are still extremely important for open source AI progress.
- zirgs 3y ago
  But you won't need to spend 50M in the future. Hardware improves all the time. Remember all those "can it run crysis?" memes? Those GPUs are now slower than mobile phones.
  - stale2002 3y ago
    Moore's law takes quite literally years to come into effect. And the AI progress time gap we are seeing right now, of "amount of time between major AI breakthroughs" is measured in weeks, not years. Going back to waiting years instead of weeks for entire technologies to be revolutionized is a major slowdown.
- jrm4 3y ago
  Right. I think you hit on the biggest question -- what is the likely MARGINAL improvements in weights that your hypothetical 50 million buys? -- even if you accept that we wouldn't be here without that first "50 million" that already went in? A reasonable argument, a possibility, is "very little." Perhaps compare to "so and so is going to put 50 million in improving the Linux Kernel?" I'm not sure this would improve anything at ALL, in fact, maybe make it worse -- given that someone will want to get something out of 50 million?
  - stale2002 3y ago
    > what is the likely MARGINAL improvements in weights that your hypothetical 50 million buys Wrong question. The marginal improvements of a specific model might be marginal, but what matters is new models for different usecases. For example, we have great image generation models 6 months ago, and one coming out now isn't going to be that much better. But, what is starting to get good right now is generative video models. And the video models are a huge marginal improvement over what was available 6 months ago. Now do the same thing for any other use case. We have reasonable good videos now, but what about 3d models? Those mostly suck right now. Lots of improvement opportunities that wouldn't exist if training new models was banned. There might be a dozen such major usecases that don't exist yet, where we can't repurpose old models, and instead need to train an entirely new model with different data. Making sure we still get that innovation is still important.

by jrm4 3y ago

ronsor 3y ago

It's pretty hard to develop an open-source project in the open if your activity is seen as illegal. We don't need AI projects lumped in with Popcorn Time and movie torrents.

jrm4 3y ago
Interesting examples you put there; FWIW you named the things that typically strongly favor the rights and desires of the consumer, despite the law, and give them a significantly superior experience to their legal alternatives (at least in the foreseeable short term.)

stale2002 3y ago

> Do you mean e.g. small "companies trying to make money?" No they mean open source projects. The overwhelming majority of all generative AI related open source projects depend on, use, or are related to publicly available weights that were trained without permission on other people's data. The entire open source AI community wouldn't exist if those weights weren't available for everyone to use for basically any reason. One could argue that the existing weights aren't going to disappear as thats impossible to enforce. But what about the updated weights? If nobody is allowed to spend 50 million dollars training a new model and then releasing that model publicly for anyone to use, without getting permission from the entire internet to make that model, well then there aren't going to be anymore new models coming out. The government isn't going to be able to confiscate everyone's gamer PCs in order to ban inference, but they can stop someone from spending that 50 mil training the next cutting edge open source model. Which means that the existing models will be all that the open source community has. Yeah they could fine tune them, but cutting edge base models are still extremely important for open source AI progress.

zirgs 3y ago
But you won't need to spend 50M in the future. Hardware improves all the time. Remember all those "can it run crysis?" memes? Those GPUs are now slower than mobile phones.
- stale2002 3y ago
  Moore's law takes quite literally years to come into effect. And the AI progress time gap we are seeing right now, of "amount of time between major AI breakthroughs" is measured in weeks, not years. Going back to waiting years instead of weeks for entire technologies to be revolutionized is a major slowdown.
jrm4 3y ago
Right. I think you hit on the biggest question -- what is the likely MARGINAL improvements in weights that your hypothetical 50 million buys? -- even if you accept that we wouldn't be here without that first "50 million" that already went in? A reasonable argument, a possibility, is "very little." Perhaps compare to "so and so is going to put 50 million in improving the Linux Kernel?" I'm not sure this would improve anything at ALL, in fact, maybe make it worse -- given that someone will want to get something out of 50 million?
- stale2002 3y ago
  > what is the likely MARGINAL improvements in weights that your hypothetical 50 million buys Wrong question. The marginal improvements of a specific model might be marginal, but what matters is new models for different usecases. For example, we have great image generation models 6 months ago, and one coming out now isn't going to be that much better. But, what is starting to get good right now is generative video models. And the video models are a huge marginal improvement over what was available 6 months ago. Now do the same thing for any other use case. We have reasonable good videos now, but what about 3d models? Those mostly suck right now. Lots of improvement opportunities that wouldn't exist if training new models was banned. There might be a dozen such major usecases that don't exist yet, where we can't repurpose old models, and instead need to train an entirely new model with different data. Making sure we still get that innovation is still important.