@windsheep also, while #China's grid is still dirtier than the West's, their architecture delivers more FLOPs per watt, ...

@windsheep also, while #China's grid is still dirtier than the West's, their architecture delivers more FLOPs per watt, which means lower token prices. Self-deploying open-weight models is the way to go because relying on USA tech is too risky. #LLM breakthrough and claimed superiority is mostly marketing spend and scale, not fundamentally better engineering. Thank God China focuses on building national infrastructure instead of optimizing ROI on token casinos. For my very modest needs I’ve canceled my Claude account. Don’t use Gemini and never touched ChatGPT intentionally. I’m fine with #mimo and #deepseek if I use non local ai otherwise it’s all running on a local server.