这篇文章以“2025 年度词汇是 slop（AI 生成的低质内容洪流）”开场，指出距离 ChatGPT 引爆全球 AI 热潮已约 3 年，但公众在 2025 年更多感受到的是垃圾信息、伪造内容与争议，而非“治病、解决气候”等承诺。作者在 2026 年初（2026-01-02 03:00，GMT+8）提出 6 个仍缺乏答案的问题，认为 AI 已进入学校、医院、招聘与政府服务等高风险场景，透明与问责变得更紧迫。

在“训练数据”上，作者强调企业以商业机密为由拒绝披露，使外界难以核实是否包含违法内容、海量版权作品或偏向英语与欧洲中心的材料；她提到欧盟计划在 2027 年年中前要求企业提供更详细的训练数据摘要，并呼吁其他司法辖区跟进。在“AGI 衡量”上，作者引用研究者观点：若问 100 位专家可能得到 100 种相近但不同的定义；尽管概念含糊，它仍被用来支撑“数千亿美元”级别投资。她也提到 OpenAI 章程的常用表述，以及据报道的“以总利润 1000 亿美元作为门槛”的内部财务指标，认为这更像可移动、易被营销利用的目标。

在监管与经济面，作者认为除欧洲外，多数地区监管推进缓慢，但对青少年、社会伤害与能源成本等担忧在上升；她建议立法者在伤害规模化前先行。关于“泡沫何时破”，她描述估值与循环投资的红旗、FOMO 延续约 3 年、以及可能的触发因素（早期用户饱和致增速放缓、强力免费开源模型压缩闭源定价）。在“钱从哪来”，她指出芯片商已获益但模型公司盈利路径更不清晰，可能转向广告等新收入；在“工作是否被取代”，她认为焦虑已出现、裁员可能扩散，政策与企业需面对潜在的大规模劳动力冲击。

The column frames 2025’s “word of the year” as “slop,” arguing that roughly three years after ChatGPT launched the AI boom, the most visible outputs were low-quality synthetic content, spam, and controversy rather than promised breakthroughs. Dated Jan. 2, 2026 (03:00 GMT+8), it lists six questions the author wants answered in 2026 as AI moves into high-stakes settings like schools, hospitals, hiring, and government services.

On training data, the author says opacity is becoming indefensible as lawsuits and harms mount, and notes the EU is set to require detailed summaries by mid-2027. On AGI, she argues measurement is unresolved: if you asked 100 experts, you’d likely get 100 different definitions, yet the term still anchors “hundreds of billions of dollars” in investment. She cites OpenAI’s charter-style definition and highlights reporting that OpenAI and Microsoft tied AGI to a financial threshold—$100 billion in total profits—calling that a dubious proxy for “intelligence.”

On regulation, she claims Europe is the main exception while other jurisdictions lag despite rising concerns (including energy costs and youth impacts). On the “bubble,” she says risk signals include lofty valuations without profits and circular funding, but that euphoria has persisted for about three years; potential stressors include slowing growth after early adopters saturate and powerful free open-source models eroding pricing power. On “where’s the money” and jobs, she argues chipmakers have already benefited while model builders must prove sustainable margins, and that labor-market anxiety is already present with more disruption likely in 2026.