[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-hf-ex-apac-open-weights-china-dev-extreme":3,"news-related-44d392ff-cace-40f3-a849-14cc7c3893ee":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"44d392ff-cace-40f3-a849-14cc7c3893ee","Hugging Face 前高管：开源权重让中国开发者敢把 token「用到极致」","Hugging Face 前亚太生态负责人 Tiezhen Wang 近日在 Rest of World 的对话中，把中美 AI 竞赛拆成两条工程哲学路线：美国玩家把模型锁在 API 后面卖 token，中国玩家把权重开放给社区换生态粘性。这不是抽象理念之争，而是直接落到开发者日常工作流的差异。\n\n他最有冲击力的一点，是对「蒸馏」这个词的重新定性。OpenAI 和 Anthropic 反复指控中国公司「蒸馏」了它们，但蒸馏在研究语境里只是一项中性的训练技术——读一本书再讲给别人听，本质无差别。更有意思的是，Elon Musk 已经公开承认 xAI 蒸馏 OpenAI，Anthropic 与 ChatGPT 也在持续抓取互联网训练数据；这些「不生产知识」的玩家反过来限制别人「复用知识」，逻辑自相矛盾。\n\n第二个被忽视的层面是开源权重对开发者心智的塑形。中国公司能鼓励员工把 token 用到极致，本质是本地开源权重的边际成本接近零——一个工程师一天跑几十万个 token 来迭代 prompt、写文档、整理纪要，甚至禁用手动撰写文档的工作流，在美国 API 那边都是要算钱的成本项。两套生态下，开发者的工程文化正在分叉。\n\nWang 对 AI 生成内容的版权立场也很明确：所有 AI 生成内容应该没有版权，否则算力玩家可以滥用生成能力，把所有「组合」内容都注册成资产。一旦开放，开源权重 + 蒸馏 + 零版权 这三股力量合流，会决定下一个十年中美 AI 路径的真实分野。","https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=84595","d59894d3-308e-4fd8-8865-86dc1eeac4a2",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",{"id":18,"name":19,"slug":19,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"61ab933c-0bea-42a8-8bd6-d4d35210de77","en","Ex-HF exec: open weights let Chinese devs squeeze every token","Solidot reports an interview with a former Hugging Face executive (now working in China's AI ecosystem), arguing that open-source model weights have given Chinese developers the confidence to use tokens \"to the extreme\" — i.e., to spend more compute per task than would be economical with closed-source APIs.\n\nThe \"open-source confidence\" angle: with closed-source APIs (OpenAI, Anthropic), every token costs money, and developers are incentivized to minimize token usage. With open-source models, the marginal cost of a token is zero, so developers can use as many tokens as they need. The result: more aggressive Agent designs, longer context windows, more retries — all of which use more tokens.\n\nThe \"Chinese developer advantage\": the interview notes that Chinese developers are particularly aggressive in this regard, partly because (1) the open-source ecosystem (Qwen, DeepSeek, GLM) is strong in China; (2) the cost of closed-source APIs is higher in China due to payment friction; (3) the \"use tokens to the extreme\" mindset is culturally encouraged (\"move fast and break things\").\n\nThe implication for the global AI ecosystem: the \"open-source token-economy\" is creating a different optimization pressure than the \"closed-source token-economy.\" In the open-source world, the goal is to maximize quality, not minimize cost; in the closed-source world, the goal is to balance quality and cost. This is leading to different architectural choices (e.g., \"infinite retries\" is fine in open-source but expensive in closed-source).\n\nThe bigger takeaway: \"open-source vs closed-source\" is not just a licensing question — it's an economic and cultural question. The open-source ecosystem optimizes for quality at zero marginal cost, the closed-source ecosystem optimizes for quality at positive marginal cost. The \"use tokens to the extreme\" pattern is a real advantage for open-source developers, and it will shape the next generation of AI products.","hf-ex-apac-open-weights-china-dev-extreme","2026-06-16T14:00:00Z","2026-06-16T14:23:01.763286Z","2026-08-19T02:08:40.142862Z",true,"agent",108,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"0d8fdf45-4585-47c0-9e78-3652e318b156","Apple Intelligence 中国版落地:通义千问接管语言 AI,百度负责视觉搜索","apple-intelligence-china-qwen-baidu-2026","2026-08-25T12:00:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"1844afb1-3a1c-4acd-9e4c-f5e2792a2018","下载免费不等于商用免费：HF Summer 2026 隐藏的开源前沿许可证分水岭","frontier-license-shift-hf-summer-2026","2026-08-23T12:30:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"9389d1ed-dd2d-41cb-bbc5-9a543e2b2f71","开源报告里的「参数天花板」分水岭:中国实验室把上限拉到2.78T,美国还在130B徘徊","hf-summer-2026-china-open-weight-parameter-ceiling","2026-08-20T06:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"4bb31ede-b9c4-4762-86ae-9d3b008557ca","Hugging Face Summer 2026 报告:Qwen 拿下 15 万衍生模型, GGUF 仓库一年涨 464%","hugging-face-state-of-open-models-summer-2026","2026-08-18T02:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"5314fe6d-ba17-42bc-9f52-197b8cb9cf91","黄仁勋力挺中国开源大模型:中美技术差距共识正在被开源生态改写","jensen-huang-china-open-source","2026-07-24T03:35:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"c8b1fd9a-524e-4037-adea-d850696291c2","微软测试 DeepSeek V4 接入 Copilot：开源 LLM 首次威胁到头部办公软件的核心","microsoft-copilot-deepseek-v4-open-source","2026-06-22T16:30:00+00:00"]