They are probably using those to train stable diffusion/dall-e/imagen like model as it's current state of the art AFAIK. As for whether google or apple do it to

They are probably using those to train stable diffusion/dall-e/imagen like model as it's current state of the art AFAIK. As for whether google or apple do it too, unless some employee makes a throwaway account and shares, we don't know. If I had to guess though, I would say google probably does, as well as publicly crawlable images. Apple most likely does too, unless they just bought the data from some other service, because they too have ml powered image capabilities and they couldn't have achieved it without obtaining a sufficiently large dataset first.