Diffusion models have become overcrowded and more focused on scaling/engineering than algorithms. I'm slowly weaning off from diffusion models, just like I did with GANs. I'm now exploring small subareas of AI that have great potential. I will have cool stuff to show in Spring.
@jm_alexia
Diffusion models (and LLMs) have brought new challenges in systems and infra never seen before in ML which, imo, is partly why it’s so interesting. The algorithms themselves are only part of the story here… unlike GANs.