AI News AI资讯 4d ago Updated 4d ago 更新于 4天前 51

Hidden Airtag reveals Amazon is trashing rare books to train AI 隐藏AirTag揭露亚马逊为训练AI而销毁珍稀书籍

Amazon operates a Las Vegas facility (VGT3) that systematically tears apart and scans rare books for AI frontier model training data An Airtag planted by a bookseller in a bulk-ordered rare book was tracked directly to this Amazon AI training facility Workers at VGT3 are trained to scan ISBNs barcodes before destroying books, supporting booksellers' theory that AI firms are methodically hunting unique printed works by working through ISBN lists Amazon declined to comment on the findings, offerin 404 Media调查证实亚马逊通过拉斯维加斯VGT3设施批量购买稀有书籍,拆毁后扫描页面用于AI模型训练 亚马逊员工在线论坛透露该设施曾因书籍供应短缺面临关停风险,后通过批量采购恢复运营 AI公司采用ISBN系统定向采购低价值但内容独特的书籍,引发藏书家对文化资产毁灭的伦理争议 亚马逊未直接回应AI训练用途,仅声明购书用于"开发改进客户服务产品" 竞争对手Anthropic和xAI公开声明不使用稀有书籍训练数据,凸显行业数据策略分化

78
Hot 热度
68
Quality 质量
72
Impact 影响力

Analysis 深度分析

TL;DR

  • Amazon operates a Las Vegas facility (VGT3) that systematically tears apart and scans rare books for AI frontier model training data
  • An Airtag planted by a bookseller in a bulk-ordered rare book was tracked directly to this Amazon AI training facility
  • Workers at VGT3 are trained to scan ISBNs barcodes before destroying books, supporting booksellers' theory that AI firms are methodically hunting unique printed works by working through ISBN lists
  • Amazon declined to comment on the findings, offering only a generic statement about purchasing books to improve customer products
  • The practice has sparked ethical debate, with critics like Michael Burry calling it "evil incarnate" while some defend it as reclaiming dormant books that sat unused for decades

Why It Matters

This is the first concrete, traceable evidence that a major AI firm is engaging in destructive book scanning at scale, confirming long-held suspicions in the rare book community. For AI practitioners, it highlights the growing desperation for unique, high-quality training data and the ethical landmines that come with it. For the industry, it raises questions about whether the race to frontier models will continue to consume cultural heritage in ways that benefit only a few closed systems.

Technical Details

  • Facility: VGT3 warehouse in Las Vegas, staffed by a team whose job is spine-removal and page scanning of rare books
  • Methodology: Workers scan ISBNs/barcodes before destruction, suggesting a systematic approach to collecting unique texts rather than random purchases
  • Supply chain: Amazon placed bulk orders through commercial channels; earlier in the year the facility reportedly ran out of books entirely, causing alarm among workers about potential shutdown
  • Data strategy: The focus on rare, untranslated, and low-distribution books indicates Amazon is targeting unique training data that competitors like Anthropic and xAI (who publicly refuse to train on rare books) are avoiding
  • Evidence method: A bookseller planted an AirTag in a rare book included in a bulk order, enabling physical tracking from purchase to destruction facility

Industry Insight

  • Training data scarcity is becoming a strategic bottleneck: The fact that a facility nearly shut down from running out of books signals that the frontier model arms race is consuming high-quality text corpora faster than supply chains can replenish them. Expect increased competition for niche, unique, and historically significant texts.
  • Ethical backlash will shape data sourcing: With prominent critics and growing public awareness, AI firms will face pressure to adopt transparent or non-destructive data collection methods. Companies that publicly commit to ethical sourcing (like Anthropic and xAI) may gain a reputational advantage even if their models are less data-rich.
  • ISBN-based systematic hunting is a new threat vector for collectors: The bookseller theory that AI firms are working through ISBN lists to ensure maximum unique coverage is now corroborated. Booksellers and libraries should anticipate continued targeting of low-monetary-value but historically significant works, creating an ethical dilemma between immediate profit and cultural preservation.

TL;DR

  • 404 Media调查证实亚马逊通过拉斯维加斯VGT3设施批量购买稀有书籍,拆毁后扫描页面用于AI模型训练
  • 亚马逊员工在线论坛透露该设施曾因书籍供应短缺面临关停风险,后通过批量采购恢复运营
  • AI公司采用ISBN系统定向采购低价值但内容独特的书籍,引发藏书家对文化资产毁灭的伦理争议
  • 亚马逊未直接回应AI训练用途,仅声明购书用于"开发改进客户服务产品"
  • 竞争对手Anthropic和xAI公开声明不使用稀有书籍训练数据,凸显行业数据策略分化

为什么值得看

本文首次通过物理追踪手段证实科技巨头系统性破坏文化遗产获取AI训练数据的行为,揭示了前沿模型竞赛中数据伦理的灰色地带。对AI从业者而言,这警示了训练数据获取方式的可持续性与合规风险,可能影响未来监管框架与行业自律标准。

技术解析

  • ISBN定向采购策略:AI公司通过扫描条形码/ISBN系统批量采购书籍,旨在覆盖尽可能多的独特文本内容,而非追求高价值首版书
  • 破坏性扫描流程:VGT3设施员工将书籍从书脊拆毁后逐页扫描,书籍在数字化后无法恢复原状
  • 训练数据稀缺性驱动:亚马逊 frontier 模型开发需要大量独特训练数据,稀有书籍提供了竞争对手未覆盖的文本资源
  • 设施运营数据:404 Media通过员工论坛讨论确认该设施曾出现书籍库存耗尽情况,后通过批量订单恢复运营

行业启示

  • 数据伦理成为新竞争维度:当主要竞争对手公开承诺不使用稀有书籍训练时,亚马逊的做法可能面临声誉风险与监管压力
  • 文化遗产保护与AI发展的冲突:AI行业需建立更透明的数据获取标准,避免以破坏性方式获取训练材料
  • 训练数据策略分化趋势:行业可能出现"伦理数据源"与"传统数据源"的分野,影响模型训练成本与合规要求

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Training 训练 Dataset 数据集 Ethics 伦理 Closed Source 闭源