I assume this is some sort of machine learning algorithm. How did they get the data set to train it on? That seems really strange. How did they test it? Seems like any engineers involved on this would come out with PTSD. Whatever they used to test it and train it with is illegal to possess and could only be used with some sort of massive government cooperation.
That's not... entirely clear actually. Well rather, what exactly the hash is based on. They call it "neural hash" which implies there's maybe some kind of embedding being used?
It's one thing to compute a simple hash. It's another entirely if they're using some type of locality sensitive hash since that very much has the possibility of false positives and would be a major invasion of privacy (which it already is).
