Y Combinator CEO Tan tells OpenAI and Anthropic 'do nothing'
Garry Tan, CEO of Y Combinator, has some pretty chill advice for OpenAI and Anthropic: "do nothing" about claims that Chinese companies are using their AI models to train rival systems.
Speaking at Demo Day, Tan suggested regulators should focus less on these disputes and more on finding the right balance between open-weight models (which anyone can use and tweak) and frontier models developed by companies such as OpenAI and Anthropic.
OpenAI and Anthropic allege unauthorized distillation
AI distillation, basically training simpler models using outputs from advanced ones, has stirred up controversy when done without permission.
OpenAI believes DeepSeek's V3 and R1 model architectures were distilled from its GPT-4 and GPT-4o models; Anthropic has similar complaints about Moonshot AI, DeepSeek, and MiniMax.
Tan says open-weight models should give people "freedom and access." He thinks regulators should pay more attention to cybersecurity and balancing open-weight and frontier models instead of restricting how AI models are trained.