Recent discussions between Nathan Lambert and Florian Brand analyze the narrowing gap between open and closed weights. They examine Kimi K3 and Qwen 3.8, focusing on how distillation affects model performance. This technical deep-dive reveals that synthetic data generation remains the primary lever for open-source progress. Practitioners should track these specific distillation patterns.