Has anyone tried this game?
Hah - I won a best-of-three 2-1 and thought I’d hang up my brain goggles at that point.
Some tips here
I managed to get to 10 correct.
Apparently, I’ve looked at a few too many StyleGan2 images & can spot most of them (well, maybe just in that limited setting - I’m guessing I’d be much less likely to spot one on its own, or it the threw in ‘both’ & ‘neither’ options on that test!)
I might be wrong, but I thin Dave Gorman (comedian) might be in the the image data set they used for “real people” (GitHub - NVlabs/ffhq-dataset: Flickr-Faces-HQ Dataset (FFHQ))
Well, the one from the dataset looks more like him than Stable Diffusions attempt, but it’s pretty consistent with the face it generates & does carry some attributes of him…
Conclusive! And I think disappointing and problematic… in fact I think I’ve seen headlines that sometimes personally identifiable information can be extracted from big networks, having been ingested by them during training.
On a lighter note: