I'm just talking about diffusion models in general and the concept of denoising. LLMs are what you would use, not the way you'd train a diffusion model but the way you'd train an LLM