@tante This also seems inconsistent with the notion that LLM output can't be copyrighted, because the machine intrinsincally has no rights. If the output is outside of the copyright regime, how can the input not be?

@tante
This reminds me of what @martinandree has pointed out in his book „Krieg der Medien“ how activists claiming to defend free speech ended up helping Big Tech to gain more and more influence.

@tante pretty sure their a fairly middle of the road libertarian group. Wouldn't call them left or right wing aligned. Don't think they have any anti-ai stances, they mainly protest mass surveillance. Just because many of the things they talk about align with left wing talking points dosnt necessarily mean that they will stay aligned with that for everything. Also I think this article is just talking about legal technicalities. Something stolen from an artist and then chopped up by an ai and blended into its training data is *technically* in the legal sense, transformative.

Now whether or not it is *moral* to condense massive amounts of artists hard work into a massive sludge of training data that attempts to replace artists in a field that very much should not have ever had AI in it in the first place is a different question.

But I don't think the EFF is talking about ethics here, their talking about legality.

@tante Tori Noble, staff attorney at the EFF, on training AI models on copyrighted works - "squarely what transformative use is." These are "general purpose tools" and "exactly the kind of creation that copyright is designed to foster".

Graham James, the Times, said they were "siding with ... A.I. companies' at the expense of American creators,". Copyright law SCOTUS said is "namely, to promote the progress of science and the arts, without diminishing the incentive to create."

EFF wtf

@tante i'm not surprised. not so long ago, I was responding to a toot by some EFF honcho trying to assert that "actually LLMs are not fancy autocomplete"

thread from there (original toot was deleted by the author)
https://mastodon.social/@conniptions/116995090045834771

@tante Yeah, came to that conclusion a while ago - they’re kind of free radicals (free for mee, not for thee)

@tante It's such a pity. They certainly seem to have changed course after Barlow's death.
Even though they've also done a lot of good things in the past, reading this one really hurts. :(

@tante This article from April Glaser in 2018 shed ALOT of light on the reasons for EFF's failures ... "The Watchdogs That Didn't Bark" ... Tip: Next time ya see @CoryDoctorow at a tech conference, don't let him get away with dodging these DISMAL PAST ERRORS + techno- #libertarian ideological positions ! #Reverse #Gringo #Centaur ! https://slate.com/technology/2018/04/why-arent-privacy-groups-fighting-to-regulate-facebook.html

@tante that’s a really disappointing shift from the EFF I used to know… or thought I knew.

@tante I have been wondering for a while if the EFF was being funded by Meta, Google, and Twitter. There position of child safety is particularly hard to ignore. It is basically "that's too bad, nothing at all can be done to improve child safety so we need to maintain the status quo."

My suspicions may be unfounded. I have met many people over the years who are single issue absolutist that are unable to thing rationally about the big picture. (For example, the "Right to Life crowd".) The EFF may manned by those sorts of people, but somehow I don't think so.

@tante I always suspected EFF. It always felt like an inside job, like a front for capitalism go-go

@tante The hard fact is that the statistical destillation process what some are calling training of LLMs is at the least partially transformative when it comes to copyright. There is precedent for that with concordances and the likes. The equally inconvenient fact is also that some of the inputs survive that process sufficiently unscathed that the result is blatant copyright infringement.