So in the past weeks we learned that decision models are apparently easy to make from traditional LLMs. What I’m curious about is whether you could get both modes from the same weights, spoken from a RAM-poor perspective.