Y Combinator CEO Garry Tan backs AI model distillation in the US

In short: Y Combinator CEO Garry Tan says US regulators should not block AI “distillation,” and he wants US open-weight AI labs to use it too.

What happened

Garry Tan, the CEO of startup accelerator Y Combinator, said regulators should “do nothing” about AI distillation, according to an interview with CNBC. He also told TechCrunch that US labs should consider doing more of it themselves.

Distillation is a way to train a new AI model by asking a more advanced model lots of questions and learning from the answers. It is like a student studying by quizzing a very smart tutor, then writing their own notes into a smaller, cheaper-to-run study guide.

Tan’s focus is on “open-weight” AI models, meaning models where the main settings, or weights, are shared so others can run and adapt them. He said smaller US open-weight labs should be able to use distillation techniques on leading US “frontier” models, which are the most capable models from major labs.

This comes after Anthropic released a report alleging some Chinese labs carried out “illicit distillation attacks.” Anthropic said these attempts involved hiding identities and using fraud or stolen login details. Tan said he is not arguing for theft, and he wants labs to be able to do this “through the front door,” meaning with normal access.

Why it matters

This debate affects who gets access to powerful AI and at what cost. If only a few companies control the best models and tightly limit how people can use them, tools could become more expensive and harder to inspect. Tan argues that since frontier models were trained on broad public knowledge, access to capable AI should be treated more like a public benefit than something locked behind strict rules.

Source: TechCrunch AI