Y Combinator CEO Garry Tan advocates for allowing U.S.-based open-weight AI labs to perform distillation on frontier AI models developed by American companies. Distillation is a process where one AI model is repeatedly queried to extract its reasoning patterns and knowledge, which helps in training new models. Tan argues that permitting this approach could diversify the available open-weight AI options in the U.S. and reduce reliance on Chinese AI models. He emphasizes that this should be done transparently without unauthorized or illicit access, contrasting with recent allegations that some Chinese labs have engaged in covert distillation using stolen credentials.

Tan's stance is notable because it challenges calls from AI leaders like Anthropic CEO Dario Amodei, who has urged regulators to crack down on distillation and illicit model extraction practices. Tan believes that AI labs should not control how users interact with APIs or limit the use of information shared through their models once accessed. He points out the irony that proprietary AI companies trained their own models on vast quantities of data, much of which was copyrighted and used without explicit permission, yet they now want to restrict distillation activities by others.

The Y Combinator chief supports a balance between frontier AI labs that push the boundaries of development and open-weight labs that provide broader access and innovation opportunities. He warns against a future where a single company monopolizes AI capabilities due to its capital advantages and talent pool, describing such a scenario as a "nightmare" and a "doomer" outcome. Tan sees healthy competition and openness as essential to keeping the AI ecosystem vibrant and accessible.

His view contrasts with other voices in the industry who warn about risks tied to rapid AI advancement and advocate for stronger regulatory oversight. While leaders like Anthropic’s Amodei and OpenAI’s Sam Altman have endorsed pacing AI progress and establishing safety standards, Tan emphasizes the importance of freedom for open-weight labs to innovate using frontier models. This debate highlights the ongoing tension between protection, openness, and innovation in the AI landscape.