I keep hearing *unsubstantiated* rumors about model optimization breakthroughs. Everyone knows that the cost of compute is too damn high.

So I’m just waiting until the next performance improvements arrive. Three years ago, a 1B param model was state of the art. Hopefully by the next year, they’ll be a model and framework which cuts the compute cost by half.

Satya playing Game of Thrones and taking the entire kingdom (and moat) in one weekend.

What UI do you use with Deepseek Coder? I need one that can format markdown text well.