That implies I’m hiding it and not proudly displaying

Later, here’s @mrose.ink musing about potato farmers and McFlurry machines with @cfiesler.bsky.social and @aramsinn.bsky.social in the charred ruins of the courtroom

This discussion really broke me

What we are seeing thus far out of at-scale AI-generated music is less major hits and more (a) comedy, (b) background muzak, and (c) pastiche that invokes direct infringement concerns. I suspect that trend will only increase as the novelty wears thin

Yes, fully agree: the rise of an AI market for any type of artistic work is very likely to spur countermarkets that further undercut the revenue of an AI market. (Cf. vinyl records.)

Or to put another way, are we really going to end up with closed loop platforms that generate outputs at industrial scale with no human inputs or users, just subscribers?

If the copyrightability problems *do* give rise to a lot of exact copying, taking that to the limit means devolving the music industry into a Muzak-producing chatbot/social network where the only revenue is access to the intermediary. Query whether that's enough of a market to cover the AI costs.

Maybe, but there are major secondarily liability, cost, and availability issues that are a long way from resolution on the use of the technology on the front end, strong political pressure from industrial publishers on distributors, and (maybe) market problems on the backend.

(For non-exact copies, the proliferation of partial AI authorship, and the uncertain line between AI tools and creation, is going to endlessly bedevil copyright infringement cases by layering patent claim-construction-style dynamics atop the already incoherent substantial similarity doctrine)

IMO, the notion that "AI generated works are not eligible for copyright protection" is true only in a formalistic sense—there are likely to be many works for which AI was used, perhaps even to a great degree, but in which *some* thin copyright still persists that suffices to go after exact copies.

Also: reminder that the underlying legal violation not being enforced is by app store and hosting companies, not TikTok itself. If the “deal” doesn’t satisfy the law, that violation continues.

Anyway, I would not be surprised to see an emergency appeal to SCOTUS and another shadow docket ruling.

Moreover, the reporting on the firing suggests that it was premised on the Copyright Office's advisory report on AI to Congress, which is arguably a legislative (not executive) function.

Unfortunately, SCOTUS's inscrutable shadow docket order in Wilcox refers in consecutive sentences to "exercise [executive] power" and "exercise *considerable* executive power," and now confronts an agency that exercises *some* executive power but also performs *legislative* (and judicial) functions.

TLDR: the judges seemingly disagree over whether the preliminary injunction equities favor the government when a fired official only exercises *modest* executive power, or rather *any* at all.

TBC, I’ve no clue what Anthropic’s actual reasons were—just that there are non-cynical reasons to settle for a non-destroy-the-company amount when you’re facing destroy-the-company liability. (Anyone who thinks fair use was a surefire win on appeal is naive or doesn’t understand copyright law.)

Maybe Anthropic was in better position to litigate up to appellate precedent than some of its competitors. But it also faced real risk! $1.5 billion is a lot, but it’s a tiny fraction of the upper end of the risk it faced.

A settlement isn’t a legal precedent. And Judge Alsup’s holding on prima facie and fair use is both not especially binding on anyone else and potentially favorable even to the extent it’s persuasive.