I'm personally happy to see effort in this space simply because I think it's an interesting set of tradeoffs (compute ∝ accuracy) - a departure from the fixed n

I'm personally happy to see effort in this space simply because I think it's an interesting set of tradeoffs (compute ∝ accuracy) - a departure from the fixed next token compute budget required now. It brings up interesting questions, like what's the equivalency between smaller diffusion models which consume more compute because they have a greater number of diffusion steps compared to larger traditional LLMs which essentially have a single step. How effective is decoupling the context window size to the diffusion window size? Is there an optimum ratio?