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

Live updates

1. 45d ago

   Garry Tan opposes regulation of AI model distillation amid Anthropic complaints

   Y Combinator CEO Garry Tan argues that regulators should not intervene in AI model distillation. This position follows accusations from Anthropic that Chinese firms Moonshot and DeepSeek stole technology by secretly diverting user requests to Claude to train their own models. Tan maintains that US open-weight AI labs should use these distillation techniques to stay competitive, while Anthropic seeks a crackdown on what it describes as illegal distillation practices.

   Why it matters

   Distillation involves prompting a frontier model to generate data used to train a smaller, separate model. This creates a conflict between proprietary model providers protecting their intellectual property and open-weight developers seeking accessibility.

   What is confirmed

   • Y Combinator CEO Garry Tan believes regulators should not intervene in AI model distillation.
   • Anthropic claims Chinese companies used unauthorized distillation of Claude.

   Still unconfirmed

   • Moonshot and DeepSeek secretly diverted user requests to Claude to train their models using millions of exchanges.

   What to watch next

   • Regulatory responses to Anthropic's call for a crackdown on distillation
   • Evidence of US open-weight firms adopting distillation techniques
   Sources used for this update (2)

   1. finance.biggo.com — Anthropic Calls for Crackdown on Illegal Distillation by China; Y Combinator Says US Open-Weight Firms Should Be Allowed
   2. theaiinsider.tech — Y Combinator’s Garry Tan Argues Against Regulating AI Model Distillation
   confidence 100%
2. 48d ago

   Garry Tan Defends Distillation as Anthropic Accuses Chinese AI Labs of Cloning

   Y Combinator's Garry Tan advocates for US open-weight AI labs to distill frontier models, arguing against restrictions despite accusations from AI giants. Anthropic claims Chinese labs Moonshot and DeepSeek secretly diverted user requests to Claude and used millions of exchanges to train their own models. While Anthropic views this as theft of technology, Tan suggests US developers should employ similar distillation techniques to remain competitive. This creates a divide between frontier model providers seeking to protect proprietary data and open-weight advocates pushing for broader accessibility.

   Why it matters

   Distillation involves using a larger, more capable model to train a smaller one. This process allows smaller models to mimic the performance of frontier systems without the same compute costs. The tension centers on whether this practice constitutes legitimate development or intellectual property theft.

   What is confirmed

   • Anthropic claims Chinese AI labs used millions of Claude exchanges to train their models.
   • Anthropic alleges Moonshot secretly routed user requests through Claude.
   • Garry Tan of Y Combinator wants US open-weight AI labs to distill frontier models.

   Still unconfirmed

   • DeepSeek secretly diverted user requests to Claude.

   What to watch next

   • Legal action or policy changes from Anthropic to block distillation
   • Public response from Moonshot or DeepSeek regarding the routing allegations
   • US regulatory guidance on the legality of model distillation
   Sources used for this update (6)

   1. Bloomberg.com — Moonshot Secretly Routed User Requests Through Claude, Anthropic Says
   2. WSJ — How Chinese AI Firms Tried to Clone U.S. AI Models
   3. CNBC — Chinese AI labs secretly used millions of Claude exchanges to train their models, Anthropic says
   4. South China Morning Post — Anthropic claims Moonshot, DeepSeek secretly diverted user requests to Claude
   5. TechCrunch — Y Combinator’s Garry Tan wants US open-weight AI labs to ‘distill’ frontier models, too
   6. CNBC — Y Combinator’s Garry Tan says 'do nothing' about distillation as AI giants accuse China of copying their tech
   confidence 90%