The ideological framing of the article seems at odds with the methodology of the research. I can kind of reconcile the idea of using an already existing model trained on unethically sourced data to make it harder to train future models without more careful data sourcing, but they also train a new model from a web scraped data set.
AFAICT, the model they produced doesn’t try to provide a substitute for the training set (and thus possibly endanger the livelihoods of the people who produced those images)
The ideological framing of the article seems at odds with the methodology of the research. I can kind of reconcile the idea of using an already existing model trained on unethically sourced data to make it harder to train future models without more careful data sourcing, but they also train a new model from a web scraped data set.
Regardless, it’s interesting stuff.
AFAICT, the model they produced doesn’t try to provide a substitute for the training set (and thus possibly endanger the livelihoods of the people who produced those images)
Any link to the source code?
I’ve been looking for some source - but its from SAND lab the makers of Glaze, but unfortunately there is no actual source code…