Popular image generation models can be prompted to produce identifiable photos of real people, potentially threatening their privacy. The work also shows that these AI systems can be made to regurgitate exact copies of medical images and copyrighted work by artists.
Here is the research paper (pdf): https://arxiv.org/pdf/2301.13188.pdf
If that is the same study as the one that made the rounds a few weeks ago than this is a bit sensationalist. They included the same images many times in the training dataset (no de-duplication) and didn’t have a very big training data-set to begin with.
If you make a “mona-lisa” AI that is trained to only paint exact copies of the Mona Lisa painting, then yes it will out-put copies of the Mona Lisa painting when you ask it to paint Mona Lisa 🤦♂️
I don’t know which study you are refering to. Is it available somewhere?
