Ben Thompson highlights a critical issue concerning U.S. open-weight model makers and their reliance on Chinese alternatives. U.S. companies are hindered by frontier labs' terms of service, making their models inferior and forcing them to distill already distilled models via Chinese intermediaries. Thompson questions the inherent negativity surrounding distillation, pointing out that LLMs themselves are distillations of internet knowledge. He argues that open-weight models foster innovation and believes frontier labs will adapt.However, dependency on China for this process is problematic. Thompson proposes U.S. legislation to declare data collection for model training as fair use and prohibit terms of service that restrict distillation for U.S. companies. He emphasizes that preventing distillation, which is essentially API querying, is practically impossible. Instead, he advocates for a proactive copyright policy that protects labs while ensuring their learnings fuel broader innovation.While acknowledging that leading Chinese models like K3 are not solely products of distillation, Thompson states it's a crucial component in their rapid development cycle. Chinese entities treat all models as open, disregarding terms of service. Western companies respecting copyright are thus forced to wait for Chinese releases, like K3. Thompson expresses skepticism towards the objections raised by OpenAI and Anthropic regarding the distillation of their models, given the paradoxical nature of the situation.