Garry Tan, CEO of Y Combinator, stated that he would not implement restrictions on AI model distillation. He argued that regulators should focus on balancing open-weight models with proprietary frontier models rather than banning distillation practices.
The debate over distillation intensified as companies like OpenAI and Anthropic accused Chinese firms, including DeepSeek, Moonshot AI, and MiniMax, of using their model outputs to train smaller systems. US government agencies have also flagged these activities as security risks.
Tan noted that the industry's moral stance on distillation is complicated by ongoing copyright lawsuits against OpenAI and Anthropic regarding their own training data. He suggested that as long as frontier models maintain performance and price premiums, open-weight models can foster innovation.
Regarding AI safety, Tan expressed skepticism toward 'AI doomsday' scenarios. He emphasized that immediate threats, such as AI-driven cyberattacks and infrastructure infiltration, require more urgent attention than hypothetical existential risks.
Y Combinator continues to prioritize the sector, with 149 out of 196 startups at its recent Demo Day focusing on machine learning and artificial intelligence. This represents over three-quarters of the total projects presented at the event.
