Re:Who's winning in Sarah Silverman's copyright suit against...
by hadlock on Sunday February 18, 2024 @11:43PM
(#39426302)
The barrier of entry to building a "foundational model" is less than $400 of compute time (72 hours) on a handful of rental A100 using widely available open source data sets. Open source projects by ICs are already on par with GPT3. It's gonna be really hard to regulate stuff in its current form, which is already commercially beneficial
Re:Who's winning in Sarah Silverman's copyright suit against...
by s1gnp0st on Sunday February 18, 2024 @10:58PM
(#39426125)
I can think of several large organizations who would love small AI companies to be pushed out of the space with a tax.
Re:Who's winning in Sarah Silverman's copyright suit against...
by hackerlight on Sunday February 18, 2024 @11:03PM
(#39426137)
No. These are profits on foundational models. Small companies playing catch up don't make profits on foundational models.
We're entering a new paradigm that copyright is not sufficient to handle. These AI models have sucked up the output of humanity itself, the entire commons. The public vs private distinction isn't that important. They're going to cause job losses across many fields, so we need a new solution. Relying on the court system to enforce copyright claims for the benefit of a single claimant is not a solution that scales to solve the actual problem here. Another thing is, foundational models are prone to monopoly because the barriers to access compute are so steep.