GPU Prices: AI is the new Crypto

Main > Everything Else

<< < (5/6) > >>

fallacy:
I knew that at the start of 2023 that AI was going to be the new buzz word for the year. Kind of like .com was or Block Chain.

Does your toaster have some of that AI in it to make it better?

Just watch this funny ass video, I don't think Nvidia is going to be any kind of AI future. Sounds like they are just trying to make some money off the Buzz word this year.

RandyT:
Here's some of that competition I talked about earlier, which could take the pressure (and profits) off from Nvidia. Per the company, their tech is Pytorch (Nvidia application) compatible, scalable with with no code rewrites, 50x to 200x faster, has considerably lower power consumption and has similar large-scale deployment costs. While this won't directly impact the requirements of smaller organizations with the need for in-house solutions, the company is embarking on multiple AI data centers in the US, and the increases in efficiency could result in lower computational costs for organizations who use remote processing. This isn't a "pie-in-the-sky" theoretical endeavor either. Apparently their tech has already been deployed in the UAE.

nitrogen_widget:
So i've read AI has been using copywrited material uploaded to the internet to help itself along and now those authors are filing against it.
will that set AI back a bit should the copywrite laws be enforced?

RandyT:

--- Quote from: nitrogen_widget on September 10, 2023, 09:27:00 pm ---So i've read AI has been using copywrited material uploaded to the internet to help itself along and now those authors are filing against it.
will that set AI back a bit should the copywrite laws be enforced?

--- End quote ---

IANAL, but in my view, virtually every bit of knowledge and inspiration that anyone on the planet currently possesses, is the result of exposure to the works of others who built upon the work of those before them. AI is "standing upon the the shoulders of giants", just as we do.

If those at the copyright office have done any research, they will understand that AI models are not just an elaborate compression scheme and any works to which the AI was exposed, are not copied verbatim into the resulting models and as such, should be covered by "fair use". If the model is "guilty" of anything, it is the act of "reading" the work a number of times and learning from it as a human might, possibly without compensating the author for doing so. If I were on a jury, I would view the damage amount to be limited to the purchase of exactly one copy of the book, if it was a work not allowed to be freely readable on the Internet. If that work was obtained from a public library, I'm not so sure even that wouldn't be excessive.

However, anything which it generates could have infringing elements of which the end user should be cautious. For example, it could instruct someone on how to build a patented invention and that, in and of itself, is not infringing activity. But if that same user does not do their due diligence to see if such a thing is already a patented invention, they would likely find themselves with a problem if they attempted to market it. On the same note, one could describe to Stable Diffusion a painting by a specific artist, and then note that artist in the prompt to get a very similar, if not nearly identical work. Obviously, it's likely that the artist would find that to be an infringement. However, if I am to understand current law, after a certain percentage of change, it is no longer considered to be an infringing work, even if a layperson might incorrectly conclude it was created by the same artist, as art styles are "borrowed" regularly and often by human artists.

But given that the patent office almost regularly accepts applications which flout the laws of physics, it may be asking too much for sane thinking to prevail at similarly situated agencies.

At the end of the day, AI on it's own is not specifically tied to any specific data. In fact, most of what makes it attractive to corporate interests is the ability for it to expound upon private data sets relating specifically to whatever that company is looking to achieve. Also, educational institutions have virtually free-reign when it comes to matters of copyright, so the demand from that sector would also be mostly unaffected. And finally, with Meta releasing their models into the public domain, they have let the proverbial "genie" out of it's bottle. So regardless of what happens next, the technology will continue to grow and become more useful to individuals. There are literally thousands of fine-tuned derivatives in the wild and more being created (and released) every day.

Tl/dr: No matter what happens by decree, the demand for AI-capable hardware will not slow down. Only new technology which surpasses the abilities of current GPU technology will slow the demand for graphics cards...and companies are tripping over themselves to create it.

lilshawn:
someone has copyrighted 70 some-odd billions of note sequences. does that mean music is dead?

Navigation

 Message Index

Go to full version