https://twitter.com/RgroseA/status/1639152385703702528?s=20
(in case twitter gets nuked the tweet is about Objaverse, a newly-released open dataset of text-paired 3D objects. It's on Hugging Face and its paper is on arxiv. The tweet I linked QRT the release and says "Just a friendly heads up, this dataset contains assets I (and tens of thousands of others) made under a CC Attribution-NonCommercial-ShareAlike license. So if anything trained on this dataset is used commercially, it will be in violation of that license.")
I once said that human beings are vibes-based creatures. We might be smart, bust mostly we do things based on vibes, and we spend quite a lot of our lives figuring out why some things have good vibes and why some feels off.
Anyway, the vibe towards generative AI (art, writing, bust mostly art) right now is rancid, especially as perceived by artists and related creatives, and for good reason. Stolen art is already abound even before the whole AI debacle, but this took it up a notch. It absolutely does not help that major organizations behind the tools to generative AI are, well, major, and they simply don't give a shit that they're taking other people's work without permission.
So I understand the knee-jerk response to this dataset being released. It'll just facilitate more art thief, won't it? And hell, I sure can't guarantee that there won't be people who do exactly that: taking this dataset, use it to generate 3D art, and then use it to gain money without giving a lick of credit back.
But bad actors have always existed, and just because something makes it easier for bad actors to do bad things, it doesn't mean that something shouldn't exist. Well, sometimes it does mean that something shouldn't exist, but my honest opinion is that this is not life-threatening enough for us to go that far.
Back to this Objaverse dataset. I noticed it only takes objects from Sketchfab licensed under Creative Commons (CC). The license specifically allow anyone to remix these 3D objects as they will, as long as they follow certain rules, depending on the version of CC (some disallows commercial uses, some force all remixes to be released under the same license).
And like, I'm pretty sure using those licensed-under-CC 3D models to then generate more 3D objects would fall perfectly fine under that license! It's not too different from a guy downloading a model and then using their application to change the model up. As long as the result is licensed according to the items that were used. If you use everything in the dataset then that means you can only use the least permissive license available.
I can think that a lot of good can come of this! Someone mentioned that it could help object detectors to adjust to more complex objects. Could help a lot of research on 3D space. It can be used for things beyond the generative AI field, even if that's what it was made in mind for.
The problem is that the vibe of the entire generative AI field is poisoned, by rampant art theft, by the general disrespect for the artists who are source of these datasets. So now many people's first thought about it is how bad actors can make use of it. It sucks.
I'm thinking of open source software and their permissive licenses, and how they've survived this long only through "good vibes" and collective trust that the licenses are honoured. A miracle when compared to this generative art AI debacle! But it does show that it's at least possible, to get by through trust alone even when bad actors are still possible.
An aspect I just thought of: software engineers tend to be more respected (read: paid well) than artists. Well, hell. That's a factor. Then again there's also an increasingly common issues of open source software being maintained thanklessly by people who can't afford it and go into debt even though their creation helped multimillion-dollar companies who don't, by the way, lend them a hand.
Late capitalism sucks.
