a16z partner Lisha Li talks with AI researcher and Trillium founder Nathan Lambert about how durable the gap between frontier labs and open models actually is. Lambert argues OpenAI and Anthropic hold structural advantages in compute, capital and the ability to pioneer new training paradigms, while open-model labs, especially in China, have proven unusually good at following close behind.
They work through how much of that comes from distillation versus talent, why compute remains the binding constraint, and how data, reinforcement-learning environments and post-training increasingly shape progress. The conversation also covers Trillium, Lambert’s new nonprofit research lab, and its plan to release RL environments and document how AI is being used to automate AI research.
