It is trivial to contrive a distinct model that behaves the same, no?
Regardless, most uses of a model won't provide direct interaction. This entire line of thought seems starved of use. Unless we're talking about LLM providers themselves... but that is a separate topic entirely
Yep. Cursor’s Composer 2 model is a good example, though it is not clear if they entered into an agreement with Moonshot before they got found out in March this year [1] or after.
Software licenses aren't enforced through litigation as much as they are enforced through the _threat_ of litigation and legal risk. In other words, pretty much every company pays lawyers to minimize legal risk. Those lawyers inevitably look at all the contracts, agreements, and software licenses, and tell the C-suite what to do in order to keep their legal exposure as low as possible. "Don't violate other companies' IP," is pretty low-hanging fruit in those conversations.
It is a very rare (and ballsy, and perhaps incompetent) company that ignores their lawyers' recommendations to adhere to the letter of all of the software licenses they are bound to.
If weights aren't copyrighted then there's no IP to license, other than the bit of software that runs the whole thing and it's probably not too hard to write that yourself. There might even be open source reimplementations soon.
Lots of companies are willing to push the boundaries of IP when there's a gray area. One example is Google's implementation of a Java API when they made Android, for which they got sued, and won their case. Another example is the AI companies training their models on copyrighted material.
> this was partially tested last year in Thaler v. Perlmutter which affirmed that copyright requires human authorship
Model weights are not authored by humans. Neither is AI output, which is why that's not copyrighted either. In another realm, a famous photo taken by a monkey was deemed by the courts to be free of copyright.