Who's afraid of Chinese models?
Points and comments are a snapshot, not live.
Chinese AI models' open weights are commoditizing intelligence, shifting advantage to lower serving costs.
Stratechery argues that fears over Chinese open-weight models like Kimi K3 and Qwen3.8 Max are overblown economically. While cheaper to develop, they incur real serving costs (COGS). The key insight is that intelligence itself is becoming a commodity; what matters is the cost to produce it. U.S. frontier labs like Anthropic benefit from supply constraints and premium pricing, but Chinese models, aided by distillation of frontier models, threaten this. China's strategy is to commoditize AI to benefit its manufacturing and robotics advantage, per Xi Jinping's speech endorsing openness.
What commenters are saying
Commenters split on distillation. One camp argues it is fair use, noting frontier labs scraped the internet themselves; as one put it, "You're trying to kidnap what I've rightfully stolen!" Another camp dismisses the notion that Chinese models rely heavily on distillation as American exceptionalism and ill-informed, noting known API-call campaigns were small-scale. A third camp debates whether governments can or should bar ToS clauses forbidding distillation, analogizing to non-compete clauses in software licenses.