Proposal to Legalize Model Distillation and Fair Use to Counter Chinese AI Models
Analyst Ben Thompson proposed that the US should pass laws establishing model training data collection as fair use and banning terms of service that prohibit model distillation. Additionally, Alibaba reversed its previous stance by releasing Qwen 3.8 Max as an open-weights model, potentially influenced by Chinese leadership's push for open-source collaboration. This proposal addresses the competitive pressure US AI companies face from Chinese open-weights models like Qwen, while highlighting the hypocrisy of labs training on unlicensed data but banning distillation. Legalizing these practices could accelerate domestic AI innovation by allowing developers to legally build upon existing models and data. Distillation involves querying a larger model's API to train a smaller, more efficient student model, a process that is currently restricted by many proprietary AI providers' terms of service. The shift in Alibaba's strategy to release Qwen 3.8 Max suggests a state-backed alignment toward open-source AI to foster global collaboration.
## BACKGROUND
Knowledge distillation is a machine learning technique used to compress models by transferring knowledge from a large, complex "teacher" model to a smaller "student" model. Open-weights models allow users to download and run pre-trained model parameters locally, differing from fully open-source AI which also provides training data and code. Alibaba's Qwen is a prominent series of large language models developed by Alibaba Cloud.