Yang Zhilin Kimi interview

بواسطة Vojtech: Yang Zhilin

The best model won't come from whoever has the most GPUs. It'll come from whoever builds the right organization around them. That's the whole argument Yang Zhilin makes, and it lands differently when you consider he runs Kimi, a Chinese AI company valued at $50B. What stuck with me is how little he talks about specific models. His focus stays on organizational structure, not the tech itself. He'll tell you OpenAI didn't win on tech. Google invented the Transformer but its structure could never have built ChatGPT. His first principle: keep improving compression and you get higher intelligence, eventually past humans. He thinks that's already settled. AGI can't be planned like an app. The org has to let things emerge, not execute a roadmap. Moonshot got its name from a Pink Floyd album. He used to be a drummer, ties it all back to rhythm. If you're trying to pick which model to actually run, there's a resource listing AI models and their costs below.