Y Combinator CEO Garry Tan holds a contrarian view regarding Chinese AI labs that employ distillation methods to learn from leading model developers—he'd rather see regulators keep their distance. Moreover, he suggests American AI companies might benefit from adopting similar tactics themselves.

"I would do nothing," he stated during a CNBC interview earlier this week. He went on to suggest that perhaps the U.S. should establish its own distillation framework, arguing that this could result in "a more robust set of open-weight options that aren't Chinese."

Distillation refers to the practice where a model developer repeatedly queries another model to understand its functionality and reasoning processes. This technique is widely and legitimately employed by AI labs for training new models. This week, Anthropic published its second report accusing Chinese labs of conducting "illicit distillation attacks"—concealing their identities to distill without authorization, and resorting to fraud and stolen credentials in the process. Anthropic CEO Dario Amodei has previously made public appeals for U.S. regulators to take action against distillation.

It's worth noting that the leader of Silicon Valley's highly regarded and productive startup accelerator holds a different perspective. To be clear, Tan isn't suggesting American AI labs should employ stolen credentials for distillation purposes. Rather, he believes they should have the freedom to enter through legitimate means.

His position actually rests on two pillars. First, he considers it excessive for AI labs to control how their customers utilize information obtained from their models. Second, he points out that proprietary AI labs never sought permission when they absorbed vast amounts of human knowledge to train their systems—they notably consumed substantial copyrighted content without obtaining consent from intellectual property owners.

Tan, who uses AI so extensively that he once characterized himself as experiencing "cyber psychosis," desires equilibrium between open-weight AI labs and frontier labs. "They are at the frontier and driving it forward. We want that to be fundable, and be a great business model ongoing," he explained to CNBC. "You want open weight models to give people freedom and access."

From his perspective, the genuine AI doomsday scenario involves all the tremendous capabilities of frontier AI becoming concentrated in the hands of one dominant, proprietary entity. "The nightmare scenario, the doomer scenario for AI is that there's just one company," he remarked. "It has the best access to capital. It has the best AI researchers. It runs away with it and suddenly there's one company that's monolithic. And that would be bad."