Y Combinator's Garry Tan says 'do nothing' about distillation as AI giants accuse China of copying their tech

• Distillation controversy involves Chinese firms like DeepSeek and Moonshot AI allegedly using outputs from advanced models like GPT-4 to train their own systems, prompting a joint advisory from NSA, CISA, and FBI
• Tan believes regulators should prioritize maintaining a price premium for frontier models while allowing open-weight models to provide freedom and access, calling it 'a tightrope' balance
• On AI safety, Tan dismisses near-term job loss fears as 'decades away' and urges focus on immediate risks like cybersecurity breaches rather than doomsday scenarios

Summary

Key Position: Y Combinator CEO Garry Tan advocates a hands-off approach to AI model distillation amid growing concerns from major AI companies about Chinese competitors copying their technology. He stated he would "do nothing" about distillation and suggested the U.S. could develop its own distillation regime instead.

The Distillation Controversy: Model distillation involves using outputs from advanced AI systems to train smaller, less capable models—sometimes illicitly. Anthropic has accused Chinese firms including Moonshot AI, DeepSeek, and MiniMax of this practice, while reports suggest DeepSeek's V3 and R1 models were distilled from OpenAI's GPT-4 and GPT-4o. U.S. security agencies (NSA, CISA, FBI) issued an advisory on the topic Tuesday.

Tan's Regulatory View: Rather than restricting distillation, Tan advocates for regulatory balance between open-weight models and frontier models, provided premium models maintain viable pricing. He favors open-weight models for providing "freedom and access," calling this equilibrium "a tightrope" worth pursuing for optimal outcomes.

AI Safety Perspective: Tan urges focusing on "science fact, not science fiction," prioritizing immediate risks like cybersecurity over extinction scenarios. He cited Anthropic's recent blocking of Claude access in five cases involving foreign scientists potentially developing bioweapons. Regarding job displacement, Tan believes impacts will unfold over decades as workers transition toward creative pursuits.

Market Activity: Despite varied AI risk perspectives, Y Combinator remains heavily invested in the sector—149 of 196 startups (76%) presenting at Demo Day were AI and machine-learning ventures, signaling continued strong investor interest in artificial intelligence.