Y Combinator's Garry Tan wants US open-weight AI labs to 'distill' frontier models, too

Key Points:

• Y Combinator CEO Garry Tan opposes regulatory crackdowns on AI distillation techniques used by Chinese labs and suggests U.S. AI labs should adopt similar practices to foster open-weight AI models domestically.
• Distillation involves prompting a model extensively to understand its reasoning, a common method for training new AI models, but Chinese labs have been accused by Anthropic of illicitly using this technique through fraud and stolen credentials.
• Tan argues that AI labs should not restrict how customers use information from their models and highlights that proprietary AI labs themselves used vast amounts of copyrighted data without permission to train their models.
• He envisions a balanced AI ecosystem where frontier labs drive innovation and open-weight labs provide accessible alternatives, warning against the concentration of AI power in a single dominant company as a major risk.
• Tan advocates for government roles in normalizing access to AI intelligence as a public good rather than locking it behind restrictive terms, promoting freedom and competition in the AI industry.