But I don’t think an occasional user being sloppy would be grounds for a lawsuit. So I suspect a scraper did indeed go haywire and downloaded porn en-masse.
Whether they trained on it is another matter. I suspect that due to the prevalence of porn on the internet there would be incentives to filter out such content (if anything, just because it’s quantity would outweigh the non-porn content) for model training.
> “[T]he small number of downloads—roughly 22 per year on average across dozens of Meta IP addresses—is plainly indicative of private personal use, not a concerted effort to collect the massive datasets Plaintiffs allege are necessary for effective AI training,” Meta writes.
Whether they trained on it is another matter. I suspect that due to the prevalence of porn on the internet there would be incentives to filter out such content (if anything, just because it’s quantity would outweigh the non-porn content) for model training.