The kids really have no idea how tenuous computing in general was back in the olden days. Some of the stability issues in the 20th century translated to modern systems would be akin to black smoke coming out of your computer if you happened to have the wrong two programs running at the same time.
They're all looking for outside money because they're all looking for outside money, and so need to keep up with their competitors investments in training. It's a game of chicken. Once their ability to raise more abates, they'll slow down new training runs, and fund that out of inference margins instead, but the first one to be forced to do so will risk losing market share.
As somebody who occasionally gets tiny ASCAP checks I think an ASCAP/BMI model might work for artists (and maybe even writers?) I guess this is more like SESAC, but maybe that's how this will end up working.
"What happens when an LLM outputs a patented algorithm?" remains a huge land mine out there, particularly since patent infringement does not require intent or even knowledge, and these models have trained on every patent ever granted.
I'm afraid as of last week this is now as settled as it gets in US law: the output of LLMs is not per se copyrightable, though arrangements and modifications of it can be. It's like a producer who made a song entirely with public domain audio samples: he can't then demand the compulsory license when someone resamples that song.
If you've been storing it with strings loose and are bringing it into tune you definitely gradually tighten all the strings rather than just doing it in sequence, for that reason
That gets tricky: To the extent that an AI is just a tool — along the lines of a trained pair of hands executing a human prompter's specific, detailed instructions — the human prompter might qualify as an "author."
From the U.S. Copyright Office in January 2025:
"The Office affirms that existing principles of copyright law are flexible enough to apply to this new technology, as they have applied to technological innovations in the past. It concludes that the outputs of generative AI can be protected by copyright only where a human author has determined sufficient expressive elements.
"This can include situations where a human-authored work is perceptible in an AI output, or a human makes creative arrangements or modifications of the output, but not the mere provision of prompts.
"The Office confirms that the use of AI to assist in the process of creation or the inclusion of AI-generated material in a larger human-generated work does not bar copyrightability.
"It also finds that the case has not been made for changes to existing law to provide additional protection for AI-generated outputs."
The closest analogy is to a music producer sampling public domain audio. The composition will be protected by copyright but each individual sample will not.
Every single CGI rendered frame of Shrek is protected by copyright because it was human authored. It they used a diffuser to make Shrek 7, the individual frames would not be protected by copyright but their arrangement into a movie could be. That's a hugely different legal situation (for instance, if I chopped it up and made my own remix of it that would be protected).
Though this is complicated by the fact that the LLM initial training may have been massively illegal (or at least massively tortious). There's still a bunch of legal shoes to drop, one way or the other
reply