AI 周刊第514期:应用型AI已到来:哪些在奏效,哪些被撤回,以及为何是现在

内容来源:https://aiweekly.co/issues/applied-ai-is-here-whats-working-what-got-pulled-back-and
内容总结:
AI应用实例库发布:159个真实部署案例揭示行业成败
今日,一份名为“AI应用实例库”的行业报告正式上线。该库收录了横跨21个行业的159个真实AI部署案例,详细记录了其中77个案例所使用的工具、供应商及最终成效。值得注意的是,有6个项目已被叫停或逆转,报告认为这些“失败案例”或许是文件中最具价值的部分。该库完全免费开放,无需注册,旨在帮助用户在投入预算或撰写方案前进行检索查询。
实际见效的领域
报告显示,AI在任务聚焦、成效可量化的场景中落地最为扎实。典型案例包括:
- OpenAI自身已大规模采用自家代码开发代理,内部使用率从去年8月的约40%飙升至97.9%,并扩展至法务和招聘部门。
- 英国国家医疗服务体系(NHS)在全国范围内使用AI读取胸部X光片,一项2000万英镑的扩建项目已帮助超过400万患者获得更快的肺癌诊断,将复杂病例的分析时间从8天缩短至4天。
- 餐饮公司Wonder部署的机器人每小时可制作500碗食品,而人类厨师每小时仅能制作约45碗。
- 搭载Momenta城市导航辅助驾驶软件的车辆已接近100万辆,应用于丰田、奔驰、比亚迪、通用和奥迪等品牌。
- Pinterest的AI广告引擎首次实现单季营收10亿美元,采用其AI驱动广告活动的广告主,投入增速几乎是未采用者的两倍。
- 中国邮政在广州分拣中心部署人形机器人,每小时处理高达1200个包裹,该站点日处理量达650万件。
企业叫停的教训
报告中的6个叫停案例被视为最“划算”的教训,揭示了AI的常见短板:缺乏机构知识、难以应对边缘情况,以及会“自信”犯错。典型实例:
- 福特公司重新雇佣了350名质检员,因其自动化质检系统产生的缺陷只能由经验丰富的人类检验员可靠捕捉。
- Waymo在四个城市暂停了机器人出租车服务。此前一辆车在亚特兰大陷入洪水被困约一小时,随后公司暂停了亚特兰大、圣安东尼奥、达拉斯和休斯顿的服务。
- Meta放弃了自动化仇恨言论审核,转向“社区注释”模式后,针对美国立法者的辱骂和种族主义帖子在六个月内增加了两倍。
- 美国一名联邦法官驳回了政府效率部(DOGE)使用AI削减拨款的行为。DOGE曾用ChatGPT将约1亿美元的国家人文基金会拨款标记为“涉及DEI”,其中竟包括一部大屠杀文学选集。法院认定此举构成违宪的“观点歧视”。
- 北卡罗来纳州维克县学区禁用AI检测工具。一名学生因检测被标红而获零分,上诉后,第二名教师认定未使用AI,成绩被改为满分。
审计时代来临:证据为王
报告指出,AI市场已从“能否做到”转向“成本多少、是否奏效”。亚马逊首席技术官沃纳·沃格尔斯公开表示,企业正在转向更便宜的开源模型,因为账单变得真实。他举例称,优步在四个月内烧光了整个2026财年的AI预算,另一家公司则在一个月内花费了5亿美元后才限制员工使用。随着买家开始审查账单,OpenAI、Meta和xAI等公司正竞相以低价抢占市场。Adobe、亚马逊、花旗等公司已被曝光正在限制员工AI使用。
当预算面临审计时,“我们认为它会奏效”已远远不够。买家希望看到谁已经尝试过,结果如何。这正是建立该实例库的全部原因——失败案例与成功案例同等重要。
使用建议
报告建议用户:优先按行业和职能筛选;在阅读成功案例前先阅读叫停案例;每个案例均附有来源链接,可直接用于预算会议或项目提案。
核心结论
- 成功案例具体但不均衡:仅在任务明确、可量化的领域真正落地。
- 失败模式相似:福特、Waymo和学区都因AI无法替代人类判断而撤回。
- 审计时代到来:预算被烧光的案例频发,在采购决策中,实例证据胜过炒作。
- 免费可检索的数据库胜过任何幻灯片:实例库专为预算审批前的调研而设计。
此外,报告还提及了Anthropic对加拿大研究机构的投资、美国联邦贸易委员会就AI准确性公开征求意见、以及纽约州一所乡村学区计划在今年秋季将人形机器人引入课堂等值得关注的动态。
中文翻译:
今天我们发布AI用例库:涵盖21个行业的159个真实、具名的AI部署案例,其中77个案例附带了所用工具、供应商和公开成果。六个项目已被叫停或推翻,而这些可能是文档中最具价值的条目。该库完全免费,无需注册,旨在让你在预算审批或撰写提案前先行查阅。本期内容将带你概览:哪些方案行之有效,哪些已被撤回,以及为何在当前阶段,一个先例文件如此重要。
【赞助商】
验证你构建、购买或集成的AI智能体。
Spec27帮助团队对自建智能体、第三方工具以及AI驱动的工作流实施一致的行为验证。实时可查证。
管理者们需要的不是更多AI新闻,而是先例:在我的行业中,是否有人真正做过这件事?用了什么工具?成功了吗?
所以我们构建了这个库。AI用例库今日上线:159个有据可查的部署案例,覆盖21个行业,其中77个附有公开成果,每个条目都链接至原始出处。
无需注册。可按行业、功能、供应商或项目是否仍在运行进行搜索。以下是该库所揭示内容的概览。
哪些方案真正奏效
当任务聚焦且成果可量化时,AI的采用是实实在在的。以下是库中的部分示例:
- OpenAI 现已在其自有的智能体上运行。97.9% 的 OpenAI 员工使用其 Codex 智能体,高于去年8月的约40%,法务和招聘部门也已接入。(来源:The Register)
- 英国国家医疗服务体系(NHS)正在全国范围内使用AI读取胸部X光片。一项2000万英镑的扩展项目已帮助超过400万名患者更快获得肺癌诊断或排除诊断,将复杂病例的分析时间从8天缩短至4天。(来源:GOV.UK)
- 一台机器人每小时可制作500碗餐食。Wonder 公司将其从 Sweetgreen 收购的制碗系统部署到了首个厨房中。而一名人类流水线厨师每小时大约制作45碗。(来源:Fortune)
- 辅助驾驶功能已应用于近百万辆汽车。Momenta 的城区导航自动驾驶软件已应用于68万至90多万辆来自丰田、梅赛德斯-奔驰、比亚迪、通用汽车和奥迪的量产车中。(来源:Reuters)
- Pinterest 的AI广告引擎首次实现单季度营收10亿美元。其AI驱动的 Performance+ 广告活动如今贡献了约30%的后端漏斗收入,而采用该功能的广告主,其支出增长速度几乎是未采用者的两倍。(来源:CNBC)
- 中国邮政使用人形机器人分拣邮件。其广州枢纽的机器人每小时可处理多达1200个包裹,该站点每日处理量达650万件。(来源:Interesting Engineering)
哪些公司收回了部署
六个被叫停的条目是库中最便宜的教训。它们展示了AI悄然失败之处:制度性知识、边缘案例,以及模型会自信犯错的一切场景。以下是部分案例:
- 福特重新雇佣了350名质检员。其自动化质量系统产生的缺陷,只有经验丰富的人工质检员才能可靠地发现。(来源:Bloomberg)
- Waymo 在四个城市暂停了无人驾驶出租车服务。一辆无人驾驶出租车驶入亚特兰大的洪水区域,被困约一小时后;Waymo 随后暂停了亚特兰大、圣安东尼奥、达拉斯和休斯顿的服务,后两个城市是出于对天气预报天气的预防措施。(来源:TechCrunch)
- Meta 放弃了自动仇恨言论审核。在切换到“社区注释”模式后,针对美国立法者的辱骂和种族主义帖子在六个月内增加了两倍。(来源:Wired)
- 一名联邦法官驳回了政府效率部(DOGE)基于AI的拨款削减决定。DOGE使用ChatGPT将约1亿美元的NEH拨款标记为“与DEI相关”。该模型将一部大屠杀文学选集也如此标记,随后一份长达143页的裁决书认定其存在违宪的观点歧视。(来源:Bloomberg)
- 维克县学校禁用了AI检测器。一名学生因被检测器标记而获得零分,后经申诉,由另一位教师确认未使用AI后,成绩被改为100分。(来源:GovTech)
审计时代,有据可查
这就是为何先例文件在本周如此重要。AI市场刚刚从“它能做到吗”转变为“它花了多少钱,效果如何”。
亚马逊首席技术官 Werner Vogels 告诉《财富》杂志,企业正在转向更便宜的开源模型,因为账单变得真实起来。他举例称,优步(Uber)在四个月内就花光了整个2026年的AI预算,另一家公司在限制员工使用前,单月就烧掉了5亿美元。
彭博社报道称,随着买家仔细审查发票,OpenAI、Meta 和 xAI 正竞相以低于 Anthropic 的价格争夺市场。404 Media 记录了亚马逊、Adobe、Atlassian 和思科(Citi)限制员工AI使用的情况,其中一家公司的月度支出飙升至1500万美元以上。
当预算开始接受审计时,“我们认为它会奏效”这种说法就不够用了。买家希望看到谁已经尝试过,结果如何。这正是我们构建该库的全部原因,也是为什么被叫停的条目与成功的案例同等重要。
如何在花钱前使用它
- 首先按你的行业和职能筛选。如果与你处境相同的人已经做过这个项目,结果就在那里。
- 在阅读成功案例之前,先看被叫停的条目。它们标明了AI仍然缺乏人类判断力的地方。
- 每个条目都链接了出处,因此你可以将原始报告直接带入预算会议或提案中。
关键要点
- 成功案例具体且不均衡。OpenAI 在智能体上运行,NHS 读取X光片,机器人每小时制作500碗餐食。AI的采用落地于任务聚焦且可量化的领域。
- 失败案例有相似之处。福特、Waymo 和维克县都在AI遗漏了人类仍能察觉的问题时收回了部署。
- 审计时代已经到来。随着 Vogels 指出预算超支、买家审视发票,在采购对话中,先例比炒作更重要。
- 免费且可搜索胜过另一套幻灯片。该库的设计初衷就是在你承诺预算之前使用它。
值得一读
- Anthropic 向包括 Mila、Vector、Amii 和多伦多大学在内的八家加拿大研究机构承诺提供1000万加元,并附赠 Claude 积分和初创公司接入权限。(来源:Anthropic)
- 美国联邦贸易委员会(FTC)就一项关于AI准确性的政策声明公开征求意见。如果你向消费者销售或部署AI,这是你需要阅读的案卷。(来源:FTC)
- 奈特-哥伦比亚大学(Knight Columbia)的《作为社会技术的AI》一文主张将AI视为一种社会技术,如果这一框架得以确立,将产生民权和监管方面的影响。(来源:Knight Columbia)
等等,什么?
- 纽约州北部的一个农村学区将于今年秋季将一个人形机器人引入课堂。位于塞内卡民族保留地的萨拉曼卡市中央学区,作为其AI和机器人课程的一部分,将部署一台 Realbotix M系列人形机器人和一个名为 Optio 的AI助教,面向约500名高中生。EdTech Innovation Hub 称这是一个里程碑式的时刻,同时指出这仍是一个单一学区的测试。(来源:EdTech Innovation Hub)
值得关注
AI从业者目前正在传阅的视频——由AI TV策划。
- 《品牌如何利用Reddit毒化AI搜索》- 404 Media
- 《桑达尔·皮查伊谈AI反弹、工作未来与谷歌新时代》- Hard Fork
本周投票
十二个月后,你所在组织的AI支出将:
上周,共有214位读者投票。
监管机构本周将其称为系统性风险。十二个月后,AI交易的走向如何?
十二个月后,你所在组织的AI支出将:
周五见。
——亚历克西斯
英文来源:
Today we're publishing the AI Use-Case Library: 159 real, named AI deployments across 21 industries, with the tools, vendors, and reported outcomes on 77 of them. Six were halted or reversed, and those might be the most useful entries in the file. It's free, no signup, and built to search before you spend a budget or write a pitch. This edition is the tour: what's working, what got pulled back, and why a precedent file matters right now.
Sponsor
Validate agents you build, buy, or integrate.
Spec27 helps teams apply consistent behavioural validation across custom-built agents, third-party tools, and AI-powered workflows.The receipts are live
Managers don't need more AI news. They need precedent: has anyone in my industry actually done this, with what tools, and did it work.
So we built it. The AI Use-Case Library is live today: 159 documented deployments, 21 industries, reported outcomes on 77 of them, every entry linked to its source.
No signup. Search by industry, by function, by vendor, or by whether the project is still running. Below is a tour of what the file reveals.
What's actually working
Adoption is real where the task is narrow and the outcome is measurable. A sample from the library:
- OpenAI now runs on its own agents. 97.9% of OpenAI employees use its Codex agent, up from about 40% last August, with Legal and Recruiting now on it too. (The Register)
- The NHS is reading chest X-rays with AI at national scale. A £20M expansion has already helped over 4 million patients get faster lung-cancer diagnoses or all-clears, cutting analysis on complex cases from eight days to four. (GOV.UK)
- A robot is making 500 bowls an hour. Wonder put the bowl-making system it bought from Sweetgreen into its first kitchen. A human line cook makes about 45. (Fortune)
- Assisted driving shipped in close to a million cars. Momenta's urban Navigate-on-Autopilot software is in 680,000 to 900,000-plus production vehicles from Toyota, Mercedes-Benz, BYD, GM and Audi. (Reuters)
- Pinterest's AI ad engine hit its first $1B quarter. Its AI-powered Performance+ campaigns now drive about 30% of lower-funnel revenue, and advertisers who adopted them grew that spend at nearly twice the rate of those who didn't. (CNBC)
- China Post sorts mail with humanoids. Robots at its Guangzhou hub process up to 1,200 parcels an hour at a site that moves 6.5 million pieces a day. (Interesting Engineering)
What companies pulled back
The six halted entries are the cheapest lessons in the library. They show where AI quietly fails: institutional knowledge, edge cases, and anything a model will confidently get wrong. A few: - Ford rehired 350 quality inspectors. Its automated quality systems produced defects that only experienced human inspectors reliably caught. (Bloomberg)
- Waymo paused robotaxis in four cities. A robotaxi drove into an Atlanta flood and sat stuck for about an hour; Waymo then paused service across Atlanta, San Antonio, Dallas and Houston, the last two as a precaution for forecast weather. (TechCrunch)
- Meta dropped automated hate-speech moderation. After it switched to a Community Notes model, abusive and racist posts targeting US legislators tripled within six months. (Wired)
- A federal judge threw out DOGE's AI grant cuts. DOGE used ChatGPT to flag about $100M in NEH grants as "DEI-related." The model labeled a Holocaust literature anthology that way, and a 143-page ruling found unconstitutional viewpoint discrimination. (Bloomberg)
- Wake County schools banned AI detectors. A student given a zero on a detector flag appealed and had the grade changed to 100 when a second teacher found no AI use. (GovTech)
The Audit Era, With Receipts
Here's why a precedent file matters this week. The AI market just moved from "can it do this" to "what did it cost, and did it work."
Amazon CTO Werner Vogels told Fortune that enterprises are shifting to cheaper open-source models because the bills got real. He cited Uber burning through its entire 2026 AI budget in four months, and another company running through half a billion dollars in a single month before it capped employee usage.
Bloomberg reported OpenAI, Meta and xAI are racing to undercut Anthropic on price as buyers scrutinize their invoices. 404 Media documented Amazon, Adobe, Atlassian and Citi throttling employee AI use, with one firm's monthly spend tripling past $15 million.
When budgets get audited, "we think it'll work" stops being enough. Buyers want to see who already tried it and what happened. That's the whole reason we built the library, and why the halted entries carry as much weight as the wins.
How to use it before you spend - Filter by your industry and function first. If someone in your seat already ran the project, the outcome is right there.
- Read the halted entries before the wins. They mark where AI still misses the judgment a human carries.
- Every entry links its source, so you can take the primary reporting straight into a budget meeting or a pitch.
Key Takeaways - The wins are concrete and uneven. OpenAI runs on agents, the NHS reads X-rays, a robot makes 500 bowls an hour. Adoption lands where the task is narrow and measurable.
- The failures rhyme. Ford, Waymo and Wake County all pulled back where AI missed something a human still catches.
- The audit era is here. With Vogels flagging blown budgets and buyers scrutinizing invoices, precedent beats hype in a procurement conversation.
- Free and searchable beats another slide deck. The library is built to be used before you commit a budget.
Worth Reading - Anthropic committed $10M CAD to eight Canadian research institutions, including Mila, Vector, Amii and the University of Toronto, with Claude credits and startup access. (Anthropic)
- The FTC opened public comment on a policy statement on AI accuracy. If you sell or deploy AI to consumers, this is the docket to read. (FTC)
- Knight Columbia's "AI as Social Technology" argues for treating AI as a social technology, with civil-rights and regulatory implications if the frame sticks. (Knight Columbia)
Wait, What? - A rural upstate New York school district is putting a humanoid robot in a classroom this fall. The Salamanca City Central School District, on the Seneca Nation reservation, is deploying a Realbotix M-Series humanoid and an AI teacher's assistant called Optio for about 500 high-school students, as part of its own AI and robotics curriculum. EdTech Innovation Hub calls it a landmark moment while noting it's still a single-district test. (EdTech Innovation Hub)
Worth Watching
The videos AI practitioners are passing around right now — curated on AI TV.
| How Brands Use Reddit to Poison AI Search 404 Media | |
| Sundar Pichai on A.I. Backlash, the Future of Work and Google’s Next Era Hard Fork |
This week's poll
Twelve months from now, your organization's AI spend is:
Last week, 214 of you voted:
Regulators called it systemic risk this week. Where is the AI trade 12 months from now?
Twelve months from now, your organization's AI spend is:
Back Friday.
Alexis
文章标题:AI 周刊第514期:应用型AI已到来:哪些在奏效,哪些被撤回,以及为何是现在
文章链接:https://news.qimuai.cn/?post=4579
本站文章均为原创,未经授权请勿用于任何商业用途