Yeah, who’s saying it doesn’t? It prevents the practices it prevents and allows the rest of the practices.
The regulation you’re going to see on this does not, in fact, prevent making LLMs or image generators, though. And it does not, in fact prevent running them and selling them to people.
You guys have gotten it in your head that training data permissions are going to be the roadblock here, and they’re absolutely not going to be. There will be common sense options, like opt-outs and opt-out defaults by mandate, just like there are on issues of data privacy under GDPR, but not absolute bans by any means.
So how much did opt-out defaults under GDPR stop social media and advertising companies from running social media and advertising data businesses?
Exactly.
What that will do is make it so you have to own a large set of accessible data, like social media companies do. They are positively salivating at the possibility that AI training will require paying them, since they’ll have a user agreement that demands allowing your data to be sold for training. Meanwhile, developers of open alternatives, who are currently running out of a combination of openly accessible online data and monetized datasets put together specifically for research, will face more cost to develop alternatives. Ideally, hope the large AI corporations, too much cost pressure and they will be bullied out of the market, or at least forced to lag behind in quality by several generations.
That’s what’s currently happening regarding regulation, along with a bunch of more reasonable guardrails about what you should and should not generate and so on. You’ll notice I didn’t mention anything about power or specific applications there. LLMs and image generators are not going away and their power consumption is not going to be impacted.
Yeah, who’s saying it doesn’t? It prevents the practices it prevents and allows the rest of the practices.
The regulation you’re going to see on this does not, in fact, prevent making LLMs or image generators, though. And it does not, in fact prevent running them and selling them to people.
You guys have gotten it in your head that training data permissions are going to be the roadblock here, and they’re absolutely not going to be. There will be common sense options, like opt-outs and opt-out defaults by mandate, just like there are on issues of data privacy under GDPR, but not absolute bans by any means.
So how much did opt-out defaults under GDPR stop social media and advertising companies from running social media and advertising data businesses?
Exactly.
What that will do is make it so you have to own a large set of accessible data, like social media companies do. They are positively salivating at the possibility that AI training will require paying them, since they’ll have a user agreement that demands allowing your data to be sold for training. Meanwhile, developers of open alternatives, who are currently running out of a combination of openly accessible online data and monetized datasets put together specifically for research, will face more cost to develop alternatives. Ideally, hope the large AI corporations, too much cost pressure and they will be bullied out of the market, or at least forced to lag behind in quality by several generations.
That’s what’s currently happening regarding regulation, along with a bunch of more reasonable guardrails about what you should and should not generate and so on. You’ll notice I didn’t mention anything about power or specific applications there. LLMs and image generators are not going away and their power consumption is not going to be impacted.