arxiv.org/abs/2310.16834
More likely, they just use this very nice work of theirs.

Wild guess: VAE-Bidirectional transformers as text embedder for per-token low dimension embeddings suitable to diffusion.

That would be an cool thing to try anyway.

Also, the whole system could already almost be seen as a form of self-improvement with some minimal human signals.

For AI to be fair and sustainable, we'd need to figure out attribution, i.e. "How much does training sample X contribute to model output Y?" Then the creator of sample X gets paid an amount proportional to what the user paid for the inference call that produced output Y.

This is ridiculous. And then people will talk about inclusivity and mental health. Sorry to speak my mind so openly, but this has to be the most toxic idea in a very long time.