[gen3]

> The development comes as regulators in the EU and US increase their scrutiny of Big Tech's investments in AI startups

Done for anti trust reasons, or because they want a bit more deniability when something goes wrong (data leak, copyright issues, NSA connections)?

Hasn't the FTC been fine with the existing structure as long as MSFT/AAPL/GOOGL/NVDA doesn't take ownership? Thats the reason why MSFT made the weird investment+chain to azure compute agreement.

[WorldMaker]

I heard third hand about some concerns that the FCC could get involved for Robocall/anti-Spam/Identity Theft/Identity Spoofing concerns and while most of the offenders are "little bad actors at the bottom", some of those concerns were that a possible starting place could be asking for stricter LLM safety precautions at the top. (An unlikely possible starting place, however, but stock markets trade on unlikelihoods all the time.)

[giancarlostoro]

I think the copyright angle is what will continue to cripple true AI since you can't ask it "hey I want like Darth Vader but red suit and slightly more like a power ranger, but still keeping a little bit of the Star Wars asthetic" it starts squaking about copyright issues, I go to a person, ask the same thing, they'll just do it.

Copyright is about blatant copying, deriving from other peoples work has been a thing since the dawn of time, and AI is crippled greatly here.

The other thing is it seems only OpenAI seemingly has scanned as much copyrighted material as they have, arguably more than anyone else. Can others do so legally? It's an entire can of worms. AI is trying to move too fast because of hype, but its not technologically there.

I say it all the time, until it runs locally its not going to truly take off. Nobody wants to fit a billion dollar bill to ask AI to write a screenplay of your uncle Tony's gangster life.

Edit:

Another angle, is the guard rails. I cannot ask AI to give me a regex of slurs so I can filter it out. (An actual positive thing.) It scolds me for even asking when it realizes it could output the nword. This would be a net good thing, and very useful! Hand crafting regex is painful. You learn it for one task, then you forget everything because you don't touch it again for months.

[c1sc0]

This is going onto be extremely unpopular, but I think an argument could be made that we need to give AI research an exemption from copyright enforcement until we have a significant lead on China in AI development. They sure as hell are not going to obey Western copyright law in training their models.

[Aerroon]

Time to move AI development to Japan?

https://www.copyright.com.au/2024/05/japan-copyright-office-...

[AlexandrB]

> The other thing is it seems only OpenAI seemingly has scanned as much copyrighted material as they have, arguably more than anyone else. Can others do so legally?

It's not even just about legality. Since everyone became aware of AI platforms are locking down to prevent this kind of scraping (without a fee) again. See: Reddit. There's also the issue of increasing quantities of AI slop that now has to be filtered through.

It almost feels like the early movers in AI are like Napster. They exploited holes in enforcement + new technology to do something that wasn't possible before and that can't be repeated in the same way either.

[giancarlostoro]

My thing with AI is, if its smart like a person, it shouldn't have any bariers, just enough of a basis to learn on the fly by using your browser if it needs to. We don't all know "everything" but we can do reasonable research to have some understanding of various topics. Anyway, I agree with you, I also think what we think is AI today is not what we will tomorrow. I also think AI should be just as free as anybody else to read any other content, as long as its not disruptive to other services.

[whimsicalism]

microsoft does have ownership stake