People (authors etc) are understandably upset about this. The models CLEARLY trained on data where they didn't own the IP. It just sort of happened too fast for laws to catch up, and now it just is what it is.
But let's assume for a moment that laws caught up. That courts decided that "no, you can't train an AI model by just scraping any content you want, and if you serve an AI model that suggests being trained on such content, you're also breaking the law".
What would happen? This seems like it would be a cataclysmic change to the economy? We'd suddenly "outlaw" several of the world's largest businesses, and the quality of AI models would jump backwards several years (Which on an AI timescale is like a century)?
But what if this is the "right" thing to do? Are we incapable of doing what's "right" now? Did the AI companies' bets on being quicker than legislation pay off?
The crime of stealing all our data has been committed but at the very least they could be forced to remove information from the training data when requested.
I don't think this is about training, it's about creating an "AI summary" specific enough for a particular query, that the user never clicks the citation link and visits the website.
Or, am I misunderstanding what is TFA is about?
> What would happen? This seems like it would be a cataclysmic change to the economy? We'd suddenly "outlaw" several of the world's largest businesses, and the quality of AI models would jump backwards several years (Which on an AI timescale is like a century)?
Yes? There's going to be a cataclysmic change to the economy when the money runs out anyway. Why should these brand new companies who skirted and ignored the law get a free pass. Is it really sensible to have environment where you can become too-big-to-fail in less than a decade because a dozen guys with cash decided it should happen?
> What would happen? This seems like it would be a cataclysmic change to the economy?
Ah we’ve entered the too big to fail era of AI corporations even before they turn a profit. How nice.
Here's the simple answer: these models were trained on the sum total of human knowledge, so they belong to humanity.
The judge rules that they are exempt from all intellectual property and trade secret laws, the labs must publish the weights and all the documentation needed to write an engine, and the labs' employees are indemnified from any adverse action for leaking anything their employers withheld.
Done.
That‘s been Silicon Valley‘s MO for about three decades
Whe should absolutely "outlaw" several of the world's largest businesses if they are making things worse for everyone else. Not doing that sends the message that the law doesn't matter if you can scale up quickly and become too big to fail. That would be an even more devastating position for society to take.