Niels AbildgaardMar 15

I have some mixed feelings on the commons, LLMs, ownership and economics. Would love some input.

I find this hard to navigate so I hope you all can extend me some grace if I mess up. Happy to read and engage, please send links. So... here goes:

I'm seeing a lot of reactions to LLM value extraction that stand on copyright, or where people are reducing their contribution to the commons as a response. This feels like throwing the game to me: the worst move in a hard situation.

#noAI #ai #llm

Show threadRaymond Neilson has moved to cosocial.caMar 15

@nielsa My position is, roughly:

  • copyright has always a dodge and a smudge, a truce of sorts, because nobody creates anything ex nihilo and disentangling the web of influence and cross-pollination is impossible, but people gotta get paid
  • as a legal construct, copyright may or may not protect any given form or piece of art, but most people aren't talking about the legal construct, but the terms of the above truce
  • ingestion is usually what's mentioned WRT copyright (see the Anthropic book settlement), but reproduction is the legal problem, and the AI companies have spent a lot of time suppressing regurgitation of the training data (to some effect but not complete elimination)
  • I think the moral case is easy to make and easy to understand, but the legal one is a minefield, and the real solution is some more explicit legislation and/or regulations
Show threadRaymond Neilson has moved to cosocial.caMar 15
@nielsa (all that said, I'm quite sympathetic to just outright banning image and video generation models entirely - they have absolutely no use case that's worth the utter shitpile they cause)
Show threadNiels Abildgaard
@delta_vee I think I broadly agree with your takes on all of these things. I always find legal solutions hard to argue about meaningfully, because I don't really respect the current global legal system. I don't think legislation will ever be the final solution - I think we should be preparing as much as we can for more structural change, rather than relying on existing (e.g. legal) structures.
Mar 16 at 7:23amWeb
Show threadNiels AbildgaardMar 16

@delta_vee Do you have any thoughts on my take on strategy? Concretely, that self-sabotaging the commons because it is exploited by LLM-creators ends up doing more long-term harm to the anti-slop "movement"? (idk which words to use to describe this exactly)

Tried to summarize my main point: https://mas.to/@nielsa/116237619916131781

Show threadRaymond Neilson has moved to cosocial.caMar 16
@nielsa Depends quite a lot on what you count as "sabotage" -- I think trying to squirrel information away is doomed, but adding more gatekeepers isn't necessarily terrible. Encouraging more highly-curated corners is probably the best bet long-term; I'm not sure there's much hope for generalist or especially universalist knowledge commons. At least, not until well after this bubble pops, and the incentives change dramatically.
Show threadRaymond Neilson has moved to cosocial.caMar 16
@nielsa I think the structural change is going to have to be to some extent within the structures of the legal system, because for all its faults, that's ultimately the mechanism we have that's anything close to both a) democratically (ish) controlled, and b) plausibly effective at restraining corporate machinery