I should be able to follow #Cara users from my Mastodon account. You can't convince me to make yet another account in a yet another silo, but I'd love to see and share more art in my home feed. Anyone have a connection to their team? Would love to chat about the #fediverse.
@KevinMarks@davew@ben I read that yesterday. From my reading of the Cara page, it is not making any promises about protecting art from ingestion into LLM datasets besides offering access to the Glaze tool. It's simply a place where generative AI slop is not allowed to be uploaded. In that sense I see no show stoppers for federation today. Artists want to be seen, makes no sense to self-isolate.
@ben@KevinMarks@davew I remember we already talked about this yesterday. I think discussions about LLM training should be happening at the legal level, not technical, unless you want DRM. Personally, I do not want to advocate for DRM. I do not believe that robots.txt is an adequate defense against LLM training, and that the onus of keeping up with all LLM user agents should not be on website authors.
@Gargron@KevinMarks@davew In offering Glaze they do seek to prevent artists’ work from being ingested. But beyond Cara, do you see the wider point about consent and author control in the fediverse?
@ben@KevinMarks@davew It's worth noting that Glaze is neither a guarantee that something won't be ingested into an LLM, nor is it something that comes at absolutely no cost; depending on setting it creates visible artifacts similar to JPEG compression that I have seen artists complain about. I don't think Glaze is relevant to the federation discussion however.
@ben@KevinMarks@davew One more thing. I am not a lawyer, but in my understanding the "default" license for any works you publish on the web is not "public domain", it is "no license". Content published on the fediverse cannot be used for anything, legally, except the intended function of the service. It is only if you want to to give more rights that you'd need "content licensing support". In that sense I also don't see this as a show stopper for federation today.
@Gargron@KevinMarks@davew Thanks for expanding. I think understanding your position on this is really important. I definitely agree that DRM is not the objective; sitting in the US, I don't know that I trust the courts to get to the right place. There's a strong possibility that the EU will do better.
@alex@Gargron I think making accounts private by default (manually approving followers, distributing to followers only) would be mostly effective. Not ironclad but a step in the right direction.
@Gargron Can they implement the required APIs to join the Fediverse without opening up to AI companies missusing these APIs to scrape off all the hosted art?
@ben@Gargron@davew Cara (and Mastodon) could add rel=license support to clarify copyright status. I wonder if we need a new license target for Machine-generated (so-called AI) images that are unable to be copyrighted.
@Gargron@KevinMarks@davew As a legal principle I think that's right (although also not a lawyer). We may need a case to act as precedent in order to truly establish this beyond doubt.
@Gargron My girlfriend just came running out of the bedroom to excitedly tell me about this new Instagram-killer named Cara. So, what serendipity you should ask! It looks very cool, and obviously we all hope it gains ActivityPub support. (Seems like they should have just stood up a Pixelfed instance.)
@KevinMarks@ben@Gargron@davew oh, sure, in the rendered web page. It's just easier for programmers to get to in the AS2 if it's a property in the JSON.
@evan@ben@Gargron@davew What we found with people serious about licencing is that an explicit link is the rendered HTML is actively useful - see Flickr for example.
@evan@alex@Gargron I think they can also federate in text-only mode (description + link to media). Not perfect UX-wise, but a much stronger protection.