Ethical AI
While I agree that the contents of the dataset is legal, I do think removing it is the best play here. I do not believe that datascraping should be illegal, but I do see their point. It's highly unregulated.
Furthermore, scraping a site that's specifically anti-AI, whether the intent was malicious or otherwise, is playing into the negative stereotype of the AI community.
So please, be mature and show them that we can be the bigger person here. Just remove the dataset and let them have this one thing.
You have a point: we shouldn't harm the open‑source community or further damage AI’s reputation (which is already pretty rough in mainstream discourse). At the same time, this kind of data-collecting is something most major labs have already done. The difference is that they usually don’t say it out loud or single out one specific community. This is more similar to what google does with its indexing.
Cara, in particular, has likely been scrapped more than once and may already exist as private data inside multiple labs, mixed in with other datasets. Such datasets will actually be images and not just URLs. There apparently was no restriction whatsoever or effort by Cara's team to prevent any activity. The "walled garden" was a fake, loosely monitored work that should not be taken lightly by its users. You cannot advocate AI-proof security if the site does not make any effort to make itself secure.
This "indexing set" I feel, is pointless - unless people can actually do something good and show what they build with it. The only pointless point being to be a reason for open debate and some questions that need answers :
- How can open-source AI advance if only elite labs get to train on massive, unrestricted data in total secrecy?
- Is "open weights" enough, or is a model truly open source only when the training data is democratized too?
- Why should a few tech giants hold absolute power over the future of AI development with government(s) help?
@CaptiveDreamer may remove it or not according to his opinions, but there's no stopping future actors to re-upload. One thing this teaches us is - once you put something online, you cannot remove it. @iscrapedcara tried. This popped up anyway. Unfortunately for us all, AI does not work without any data to train it.
P.s, Doxing, harassing HF staff and the page maintainers, throwing insults, threats for extremities, all of it is not needed by any side.
I agree with you both here, but there is one issue from which I found that stems within the community of Cara. It is a free project. Whatever issues that lie might take sometime to fix due to this being a volunteer basis.
@captivatedreamer
Industry giants have always had the upper hand over regular people. That's not new and there's nothing we can do about it. But it's not an excuse to follow in their footsteps. We can still scrape, but we must do it to sites who are either indifferent to scraping or outright encourage it.
Also, from what I've gathered, the Cara team seem to be a small independent group that built the site and maintained it out of their own pockets plus some funding from the site community. So they're not to blame here.
I'm an advocate for AI, but I also firmly believe that AI is a tool and not a solution. Artists can use AI to critique their work so they know where and how to improve. Authors can use AI to improve pacing and plots. Musicians can use AI to give them a starting beat to work off of. But the vast majority of people won't think art that's generated by AI is impressive. The vast majority of people won't buy a book if the whole story was written by AI. The vast majority of people won't listen to music the was entirely made by AI.
AI can help, but it's not the complete answer.
I'm getting off topic. If @CaptiveDreamer chooses to delete the dataset, then it's done. We give them what they want and move on. But keeping the dataset up is detrimental for both sides. It's a stain on the reputation of AI supporters and will further drive people away from AI.