Who’s Afraid of Chinese Models?

Explain Like I'm Normal

Ben Thompson argues the US should legalize training on all data as 'fair use' while simultaneously outlawing terms of service that ban model distillation. This would stop labs from being hypocritical about data usage and allow open-source models to catch up to proprietary giants and Chinese competitors by learning from their outputs. The goal is to maximize innovation speed by embracing the fact that digital secrets are impossible to keep.