天国的阶梯
Inside Asia’s oldest private library, guarding ancient scrolls in the age of AI_我的网站
一 | 【TechWeb】7月20日消息，近日，一场围绕中国AI公司月之暗面（Moonshot AI）旗下模型Kimi K3的言论风波，在海外科技圈持续发酵。事件的中心人物是刚刚上任仅两周的OpenAI“战略未来”主管迪恩·W·鲍尔（Dean W. Ball）。他在社交平台X上的一番先扬后抑、并带有强烈政治色彩的评论，不仅激起了技术社区的强烈反弹，更引发了关于中美AI竞争与监管干预的深层讨论。
Huang Gang is in his fifties. Every day, he sits before a computer screen at Tianyige Museum (Tianyige), flipping through ancient books - digitally.
As director of the information technology department at Tianyige Museum in Ningbo, East China's Zhejiang Province, home to Asia's oldest surviving private library and one of the world's three oldest remaining family-run libraries, his job title may seem far removed from old books. Yet his work centers on them: converting brittle, thread-bound antique volumes into searchable, zoomable digital files accessible on screen.
For years, Tianyige has carried out digitization in batches guided by its mission: rescue-oriented preservation, scholarly organization and public-oriented outreach. Its teams have scanned more than 3.4 million pages of ancient texts. The institution rolled out its Ancient‑Book Digital Service Platform back in 2009. By 2025, the platform had drawn over 4.6 million total visits.
To outsiders, this might look like a headlong rush into digitization. For Huang and his colleagues, it is delicate work balanced between fast‑evolving artificial intelligence models and the reverence owed to fragile paper relics. "Technology is a tool, ancient books are artifacts, and reading is the ultimate purpose," Huang told the Global Times.
Beyond tech boundaries
Deep in the storage vaults of Tianyige, tens of thousands of ancient volumes rest in silence. Some pages are curled and fragile; others bear faded script -the accumulated marks of centuries. Vulnerable to aging, wear and external damage, these precious paper relics can hardly sustain long-term repeated manual consultation, making digital preservation an indispensable way to inherit ancient book resources.
Huang's daily work involves leading his team in the digitization of the library's entire collection of ancient books, paintings and cultural relics, bringing rare treasures out of the shadows and into public view.
AI now serves as a core auxiliary tool for ancient-book digitization and research. Huang's group initially built character-recognition frameworks based on optical-character-recognition technology. By cross-checking outputs across different models, ancient Chinese characters, damaged script and blurry handwriting.
According to Zhejiang Daily, by August 2025, the Tianyige's digital platform had added over 3,700 new digitized book sets, lifting its total digital holdings past 20,000 volumes.
Digitization is never a purely technical exercise; in-depth scholarly expertise in ancient texts remains indispensable. Huang frequently collaborates with Zhao Xuteng, a museum librarian specializing in classical literature. Zhao is responsible for selecting which volumes to digitize, evaluating editions, studying content, and identifying works of significant academic or public value. While one team focuses on the technical execution, another digs deep into the textual substance, ensuring that cold digital capture remains faithful to the historical and bibliographic logic of the originals.
Staying true to reading
When he is not writing code, Huang keeps asking one fundamental question: What should technology actually do for people? "All technology ultimately serves the act of reading itself," he said. "It can never replace that."
With the rapid rise of large language models, demand for high-quality text corpora has surged. Overseas, some AI companies have reportedly purchased millions of printed books in bulk, scanned them and then shredded the originals.
"Machines can efficiently compile documents, retrieve sources and generate explanatory notes. Real engagement with ancient texts, however, is far more than mechanical assembly of characters," Huang noted.
Both Huang and Zhao stressed that human beings sit at the heart of cultural inheritance. People absorb wisdom from old texts, draw on personal experience and scholarship to reflect, and produce new knowledge - an intellectual process no AI can fully replicate.
The Global Times learned that every digitization workflow at Tianyige follows strict closed-loop protocols: collection checkout logging, noncontact scanning, AI-powered quality checks, precise image stitching and compliant online publication. The full procedure minimizes physical wear on original artifacts.
"A book is, first and foremost, a cultural relic that carries a thousand-year legacy," Huang said. "Only secondarily is it a text to be read." That principle, he added, is shared by all guardians at Tianyige.
"Media change, technologies evolve, but the essence of reading has never changed," he stressed.
Today, drawing on its comprehensive digital collections, Tianyige hosts monthly exhibitions of rare texts and organizes family reading events where parents and children explore the classics together. For Huang and his colleagues, the most gratifying reward is watching more and more ordinary people develop a love for ancient reading and a deeper appreciation for the traditional Chinese culture.
In an era swept by artificial intelligence, Tianyige's guardians tend to fragile paper relics while leveraging modern technology, preserving the genuine power of reading. More than 450 years of scholarly heritage flows onward, passing between human fingertips and glowing digital screens.
。 该图片疑似使用了AI生成技术，请谨慎甄别 鲍尔在最初的帖子中，首先承认了Kimi K3的技术实力。他将其描述为“一个非常优秀的模型”，并表示其展现出的能力“不能简单归因于蒸馏”（指从大模型压缩知识的技术），在某些应用场景下，Kimi K3的表现已接近全球顶级水平。 然而，赞誉之声迅速转向了严厉的批评。鲍尔的核心论点在于，开放权重（Open-weight）的模型本质上是一种“减速主义（decelerationist）”行为。他声称，这种开源共享模式会削弱商业AI公司的投资回报，进而阻碍整个行业所需的巨额资本投入。 不仅如此，鲍尔还对中国允许如此强大的模型保持开源表示“惊讶”，并借机渲染了AI技术可能从实验室“逃逸”的安全风险。 献计“监管恐慌”引爆舆论 真正引爆舆论的，是鲍尔提出的应对建议。他公开提议，美国政府应采取更为激进的干预手段，通过制造“监管恐慌”来遏制中国AI模型的传播。具体而言，他建议美国政府发布模糊的、暗示“可能存在后门”的软性法规，以此在市场中制造不信任感，从而从商业层面打压Kimi K3等中国开源模型的国际竞争力。 此言一出，立刻在X等社交媒体上激起千层浪。 批评者们尖锐地指出，鲍尔的建议本质上是利用国家监管机器为像OpenAI这样的美国商业巨头扫清障碍，是典型的“监管捕获（regulatory capture）”行为。白宫AI顾问大卫·萨克斯（David Sacks）也公开对此提出了质疑，认为这种将商业竞争直接与政治监管挂钩的论调非常危险。 面对巨大的舆论压力，鲍尔于当地时间7月19日晚间紧急发帖澄清。他强调，自己加入OpenAI“不到两周”，此前的言论仅为个人对未来趋势的冷酷分析预测，并非代表OpenAI的政策倡议或立场。他试图为自己的激烈措辞做出辩解，但争议并未因此完全平息。 开源与闭源路线之争的白热化 此次风波并非孤立事件。它本质上是中美两国在AI发展路径上“开源共享”与“闭源垄断”之争的又一次激烈碰撞。 月之暗面7月16日上线推出的Kimi K3模型拥有 2.8 万亿参数，在 Frontend Code Arena 全球AI大模型榜单中，Kimi K3以1679分超越Claude Fable 5，位居第一。
二 | 其完整模型权重将于7月27日向社区开放。届时，全球开发者都可下载运行这套2.8万亿参数的国产大模型。
三 | Kimi K3作为中国大模型领域的代表性产品之一，其开源策略被许多人视为对抗美国巨头技术壁垒的重要方式。而鲍尔的言论则暴露了美国部分科技精英和战略人士的焦虑，他们不仅担忧技术被赶超，更担忧开源模式从底层瓦解闭源商业模式的根基。 鲍尔虽以“个人观点”紧急降温，但这场由新上任高管点燃的争论事件也标志着，全球AI领域的竞争已不仅限于算力、算法和数据的比拼，更延伸到了地缘政治、监管政策和国际舆论的复杂博弈之中。
四 |
Current article:http://wd1b.diuhuairandaikaomangshangnen.buzz/wcxq/zf1s.html
Published on:00:00:00