Keshigeyan Chandrasegaran and Kyle Sargent release GPIC, a fully permissive 100 million-image dataset for training visual models · Digg