Comment by nunez
1 day ago
Finally glad to see big name politicians rally around this. That this is a bipartisan effort was extremely surprising to see.
1 day ago
Finally glad to see big name politicians rally around this. That this is a bipartisan effort was extremely surprising to see.
This would basically grant facebook and google a monopoly on AI -- as they'll put training on your material as part of their TOS and then be the only players with enough market power to get adequate amounts of training material.
They _already have_ a monopoly on it, by design.
The data is definitely a critical piece, but they are the only companies with the cash, hardware and talent to train frontier AI models from scratch. (The models that are fine-tuned by everyone else, to be clear.)
I don't see that changing either; there is no incentive to make training cheaper and more accessible.
I was hoping that this bill would make it possible to _retroactively_ seek legal action for copyrighted data in data sets, but, yeah, as journaled here, this will amount to a clause on an optional-but-not-optional EULA to give them "permission" to do what they were already doing, perhaps even more flagrantly.
It would grant China an even bigger victory since China's models do not have to abide by any US copyrights.
Google and Microsoft are using proxy companies to steal all the copyrighted content ever produced, and you're blaming China, or suggesting it'd be worse if they did it? Right.
1 reply →
Indeed, and perhaps not just China.
The US started off not acknowledging foreign copyrights for a long time-- until it had a large enough base of material it wanted reciprocally protected.
If not adopting these rules grants you the ability to produce SOTA AI's while most of the US can't we can expect it to be widespread.
This actually gives me a little hope-- the US cutting it's own throat this way vs other countries would be better than granting google and facebook a monopoly.