Analysis
Y Combinator CEO Garry Tan is publicly advocating for unrestricted model distillation among US open-weight AI labs, TechCrunch reported. Tan wants smaller American open-weight labs to use the same distillation techniques on US frontier models that Chinese labs have used on Western ones, arguing it would give the US a more robust set of open-weight alternatives that aren't Chinese-controlled.
Tan frames frontier models -- built on what he calls the 'bedrock of public human knowledge' -- as a form of public good, arguing that clamping down on distillation, the process of training smaller models to mimic larger ones' behavior, would hurt the broader ecosystem more than it protects any single company's intellectual property.
“Tan's position directly counters recent appeals from proprietary model developers, including OpenAI, Anthropic and Google, seeking tighter restrictions on the practice.”
His comments land days after a joint advisory from the NSA, FBI and CISA flagged large-scale model-distillation efforts by Chinese companies -- the same distillation dynamic Anthropic's own September threat report flagged when it disclosed detecting Chinese distillation attacks against Claude. Tan's position directly counters recent appeals from proprietary model developers, including OpenAI, Anthropic and Google, seeking tighter restrictions on the practice.
Tan's YC portfolio includes numerous companies building on top of both open- and closed-weight models, so his position isn't disinterested: cheaper, US-made open-weight alternatives to frontier models directly lower the compute bill for YC's own application-layer startups. Whether any US lab actually takes up his suggestion, rather than continuing to lobby for restrictions targeted specifically at Chinese distillation, is the open question this debate leaves unresolved.