I know that my link tried to explain that there's uniqueness in diffusion models - it supports the point that that's still where we are with diffusion models when that train left the station many years ago regarding LLMs.
Training the diffusion model involves learning to create what someone else already knows to recognize as a "cow". Converging to what's in the training data is not a negative.
mbpaz (npub1xcd…hnp8) Chad McCullough (npub1ded…kves) elementary (npub16nm…cc46)
