Piracy: ꜱᴀɪʟ ᴛʜᴇ ʜɪɢʜ ꜱᴇᴀꜱ

It is possible that merely training an LLM on material might be fair use. It's basically the same thing are scraping and then doing statistics. That said, I believe the burden of proof is that a use falls into the fair use exemptions; AI training might not satisfy the criteria the courts have previously established. The U.S. copyright office has published guidelines that indicate training does implicate the right of reproduction, so it if it's NOT fair use and unlicensed, it is infringement / violation.

But, distributing the output can still be a copyright violation not covered by fair use exemptions, even if training is covered by fair use. The U.S. copyright office has indicated that distributing an output could implicate various rights based on the training data that it is similar to.

In any case, LLMs (or any non-human thing) can't be given a copyright, and the U.S. copyright office says that prompts are NOT sufficient creative effort to establish a copyright on what a generative AI model outputs in response to a prompt. So, it could be that the only rights holders that get to affect the output are the rights holders of the training data.

If this is the same case that came through yesterday, the Trump DOJ brief is basically just an amicus brief. There's some sort of jurisdictional / standing issue there that I don't understand; they aren't the ones arguing in court.

Maybe a world without copyrights would just be better. But, if Disney gets protections, I want protections on my AGPL code and I want the artists enjoy to get protections as well, even if they haven't joined a licensing cabal. If that means AI is doomed, fuck it.