With quickly improving diffusion language models and rumors that OpenAI’s new image model is auto-regressive, it will be really funny if we got the optimal models for language and images backwards. Though maybe both are just equally compute and data bound.
