Most of these companies are just arguing that they shouldn’t have to license the works they’re using because that would be hard and inconvenient, which isn’t terribly compelling to me. But Adobe actually has a novel take I hadn’t heard before: they equate AI development to reverse engineering software, which also involves copying things you don’t own in order to create a compatible thing you do own. They even cited a related legal case, which is unusual in this pile of sour grapes. I don’t know that I’m convinced by Adobe’s argument, I still think the artists should have a say in whether their works go into an AI and a chance to get paid for it, but it’s the first argument I’ve seen for a long while that’s actually given me something to think about.
For those of you who didn’t read the paper, the argument they’re making is similar to Godel’s Incompleteness Theorem: no matter how you build your LLM, there will be a significant number of prompts that make that LLM hallucinate. If the proof holds up then hallucinations aren’t a limitation of the training data or the structure of your particular model, they’re a limitation of the very concept of an LLM. That doesn’t make LLMs useless, but it does mean you shouldn’t ever use one as a source of truth.