Perhaps someone will (or maybe it has already been done) figure out a business model for selling access to curated datasets that are known not to include a bunch of additional ML generated noise.
Although, to some extent I wonder how much it matters. If we're creating images using AI tools, and then sharing the best results, doesn't that become valid training data? In some sense are we supervising the learning?
>If we're creating images using AI tools, and then sharing the best results, doesn't that become valid training data?
Maybe in cases where those results are at least as good as the real thing. But in general, something being the best of some set of options doesn't imply that it's good, let alone perfect.
And besides, people will also share comically bad results.
Although, to some extent I wonder how much it matters. If we're creating images using AI tools, and then sharing the best results, doesn't that become valid training data? In some sense are we supervising the learning?