谁在害怕中国模型？

背景与摘要

本文讨论了当前人工智能模型面临的政策与开源策略的转变。作者首先引述了 Ben Thompson 提出的针对美国版权政策和模型蒸馏（distillation）的立法改革建议，呼吁通过公平竞争来促进创新。随后，文章探讨了阿里巴巴在国际地缘政治背景下发布的拥有2.4万亿参数的 Qwen 3.8 Max 开源模型，并通过一个生动的鹈鹕骑自行车的矢量插图生成案例，展示了该模型在创作过程中有趣的内部推理决策。

Summary: A discussion on Ben Thompson’s proposal for U.S. copyright reform regarding AI model training and distillation, Alibaba’s open-weights release of the massive Qwen 3.8 Max model following geopolitical shifts, and a delightful look at the model's reasoning process while generating vector art.

Via John Gruber

Summary: A discussion on Ben Thompson’s proposal for U.S. copyright reform regarding AI model training and distillation, Alibaba’s open-weights release of the massive Qwen 3.8 Max model following geopolitical shifts, and a delightful look at the model's reasoning process while generating vector art.

Via John Gruber

🏛️ Policy, Distillation, and Fair Use

在最近的一篇 Stratechery 文章中，Ben Thompson 强调了顶级 AI 实验室的虚伪：它们通过服务条款禁止模型蒸馏（distillation），尽管它们自己也在使用未经许可的数据训练其模型。为了创造一个公平的竞争环境并帮助美国的开源模型与全球同行竞争，Thompson 提出了一个清晰的立法解决方案：

In a recent Stratechery article, Ben Thompson highlights the hypocrisy of top AI labs that outlaw model distillation via terms of service despite having trained their own models on unlicensed data. To level the playing field and help U.S. open models compete with global counterparts, Thompson proposes a clear legislative solution:

“美国应该通过一项法律，(1) 明确规定为了训练模型而收集数据属于合理使用，(2) 至少禁止美国公司在服务条款中规定禁止蒸馏。停止蒸馏——官方而言这实际上仅仅是查询 API——几乎是不可能的；美国应该反其道而行之，推行新的版权政策，既保护这些实验室免受赔偿责任，又能保证他们所学到的知识能够激发其他所有人的进一步创新。”

"The U.S. should pass a law that (1) makes explicit that collecting data for training models is fair use, and (2) bars terms of service that forbid distillation, for U.S. companies at a minimum. Stopping distillation — which is nearly officially just querying the API — is nearly impossible; the U.S. should go the other way and lean into a new copyright policy that both indemnifies the labs and also guarantees that what they learned fuels further innovation for everyone else."

🇨🇳 The Shift in Chinese Open Source Strategy

Thompson 还推测了阿里巴巴的战略逆转。在五月份选择不发布 Qwen 3.7 Max 之后，阿里巴巴出人意料地发布了作为开源权重模型的 Qwen 3.8 Max。

Thompson also theorizes about Alibaba’s strategic reversal. After choosing not to release Qwen 3.7 Max in May, Alibaba surprised the industry by releasing Qwen 3.8 Max as an open-weights model.

这种转向可能深受最近一次公开讲话的影响，在讲话中提到：

This pivot may have been heavily influenced by a recent speech by Xi Jinping, in which he stated:

“我们要抓住这个千载难逢的历史机遇，鼓励开源、开放、合作与共享。”

"We should seize this rare, historic opportunity to encourage open source, openness, collaboration and sharing."

🎨 Inside the Mind of Qwen 3.8 Max

Qwen 3.8 Max 是一个庞大的 2.4 万亿参数模型（几乎与拥有 2.8 万亿参数的 Kimi K3 匹敌）。为了展示其能力，Simon Willison 分享了一幅完全由该模型生成的鹈鹕插画：

Qwen 3.8 Max is a massive 2.4-trillion-parameter model (nearly rivaling the 2.8T Kimi K3). To showcase its capabilities, Simon Willison shared a pelican illustration generated entirely by the model:

生成的描述：
一幅扁平的卡通矢量插图：一只长着橙色大喙和喉囊的白鹈鹕骑着一辆红色的自行车，它的橙色脚蹼踩在踏板上；背景是浅蓝色的天空，右上角有一轮黄色的太阳，左上角有一朵白云；自行车后面有水平的运动线条，底部是淡绿色的地面。

Generated Description:
Flat vector cartoon illustration of a white pelican with a large orange beak and pouch riding a red bicycle, its orange legs on the pedals, against a light blue sky with a yellow sun top right and a white cloud top left, with horizontal motion lines behind the bike and a pale green ground strip at the bottom.

[ Illustration Placeholder: White pelican riding a red bicycle ]

也许最吸引人的是，窥探模型大量的推理追踪记录，揭示了在幕后做出的细致入微的创造性决策：

Perhaps most charmingly, a peek inside the model's extensive reasoning trace revealed the meticulous creative decisions being made behind the scenes: * “可以加个头盔吗？不了。” * "Could add helmet? No." * “或许加个小铃铛？不了。” * "Maybe add small bell? no." * “可能需要在篮子里加条小鱼？没必要。” * "Need maybe add small fish in basket? Not necessary."

🏷️ Tags

ai | generative-ai | llms | training-data | qwen | pelican-riding-a-bicycle | ai-ethics | llm-release | ai-in-china