According to CoinMeta, former researcher Andrew Ho has resigned to start his own business, establishing a new company that will produce high-quality reinforcement learning data for large models. The initial products will focus on biology and statistical reasoning. He believes that while large models perform exceptionally well in the field of programming, they still require human intervention and lack suitable training tasks. Ho questions the trillion-dollar valuation of cutting-edge AI laboratories, arguing that even if the gross margin of their reasoning business reaches 80%, an annual revenue of 100 to 20 billion dollars would be needed to support such a valuation. He is cautious about the full potential of AI and believes it may take more than 20 years for them to truly penetrate various industries. Ho remains optimistic about the long-term growth of the AI industry but is unwilling to invest in OpenAI or Anthropic at the current valuation.