Adobe's 'Ethical' AI Tools Used Rival AI Images for Training

So let’s focus very clearly on what the reporting unearthed here, what the new part is, and I think it’s that we learned a lot more about how Firefly was trained, but it’s different to what Adobe had originally told us. Precisely so Adobe has spoken a lot, very publicly about how Firefly is really different from these rival image generators. And it turns out that it included a bunch of AI generated images, some of which are from Mid Journey in the training of Firefly. This was something I uncovered in my reporting. Brody and I worked together to pull this story together and and found that the company had done this. I mean, it was clear. It was clearly done on purpose. It is a. Huge amount of its training data, but it’s in there nonetheless. What’s important is, of course, that this was deemed the ethical version of AI that you’re paying for. And there’s been some murkiness, to say the very least, in some of the other foundational models and generative models that we’ve seen. I’m interested as to what Adobe said in response, because I’m sure they’re aware of some of these synthetic images that they were bringing in. Is there any risk of IP, this one? Adobe was totally aware of what it was doing. I mean, we were looking at Discord groups that the company runs. It was talking. It had people from the company talking on discord groups about including these images. They paid out a bonus in September to all the people whose images were used to train the first version of Firefly. And that included people such as one of my sources who is in the story, who only contributed AI generated images. To Adobe Stock, and most of those were Mid Journey made with Mid Journey.