TIFFANY HSU2026年8月18日The worst of artificial intelligence slop has cluttered Google searches with recipes for glue-topped pizza, Amazon listings with sham biographies and Facebook feeds with images of Shrimp Jesus.人工智能最糟糕的垃圾内容已经用胶水披萨食谱淹没了谷歌搜索，用虚假传记塞满了亚马逊商品列表，用大虾耶稣图片充斥于Facebook信息流。The digital sewage heap has sprawled across tech platforms, cheapening meaningful content and polluting the information ecosystem, as chatbots and video generators continue to advance.这片数字污水池已经在科技平台上蔓延开来，令有意义的内容变得廉价，污染了信息生态系统，而聊天机器人和视频生成器仍在步步推进。Now, Silicon Valley wants to take out the trash.现在，硅谷想要清除这些垃圾。Spotify, a music streamer, said it removed 75 million bulk uploads, duplicate songs and other “spammy” tracks last year — a sizable chunk of its total portfolio. LinkedIn said in July that “A.I. slop is a top priority for all of us” and developed a button for users to report it. Researchers from YouTube’s parent company, Google, described how a video service wiped 130,000 channels of low-quality content off its platform over six months.音乐流媒体平台Spotify表示去年下架了7500万条批量上传、重复歌曲和其他“垃圾”音乐——这占其总曲库的相当大一部分。领英在7月表示，“应对AI垃圾内容是我们所有人的首要任务，”并开发了一个供用户举报的按钮。YouTube母公司谷歌的研究人员描述了该视频平台如何在六个月内清除了13万个内容质量低劣的频道。领英最近为用户开发了一个“看起来像AI垃圾内容”的按钮。 LinkedInAcross the internet, tech companies are undertaking a generative waste detox, overwhelmed in many cases by the same slop first made possible by their own A.I. tools. Mobile messaging apps, restaurant review sites, scholarly archives and dating services are all rooting out or downplaying junk generated by A.I. — many relying on the same technology to pull off the purge.在整个互联网，科技公司正在进行一场生成式废物大扫除，在许多情况下，它们已被自家AI工具最初所催生的同样垃圾内容所淹没。移动消息应用、餐厅点评网站、学术档案库和婚恋服务都在根除或降级AI生成的垃圾内容——其中许多依赖同样的技术来完成这次清理。People are tiring of A.I.’s effect on the environment and its pressure on jobs and mental health. A.I. slop is just another “festering wound,” said Aidan Walker, a content creator and researcher studying online trends for the Carnegie Endowment for International Peace, a global affairs think tank.人们已经开始厌倦AI对环境的影响、对就业和心理健康造成的压力。全球事务智库卡内基国际和平基金会研究员、内容创作者艾丹·沃克表示，AI垃圾内容只是又一个“溃烂的伤口”。“Users have gradually gotten more and more upset about it,” he said. “The climate of opinion has now recognizably shifted, so these companies are finally confronting these problems.”“用户渐渐对此越来越不满，”他说，“舆论氛围已经明显转变，所以这些公司终于开始正视这些问题。”Generative A.I. can produce increasingly realistic images, audio and videos, but only some of it qualifies as slop. Tech companies are largely targeting the kind of low-grade clickbait that is easy to make, spread, consume and forget — the fast fashion edging out couture, the pink slime subsuming the steak.生成式AI能够制作出越来越逼真的图像、音频和视频，但其中只有一部分算得上垃圾内容。科技公司主要针对的是那种低质量的点击诱饵——容易制作、传播、消费和遗忘——就像快时尚挤掉了高级定制，粉红肉渣取代了牛排。Unchecked, the synthetic swamp online could “crowd out human creativity, corrode cultural value and destabilize public discourse,” European researchers wrote last year.如果不加遏制，网上的合成沼泽可能会“排挤人类创造力，腐蚀文化价值，并破坏公共话语”，欧洲研究人员去年写道。Slop has become so prevalent that Merriam-Webster, the dictionary publisher, chose it as its word of the year in 2025. Deezer, a French streaming service, said this spring that nearly half of the tracks updated to its system each day were artificially generated, many of them never heard by users. A.I. generates more than a third of what is uploaded to Apple Music.AI垃圾内容已经变得如此普遍，以至于词典出版商韦氏公司在2025年将其选为年度词汇。法国流媒体服务Deezer今年春天表示，每天更新到其系统的曲目中近一半是生成的，其中许多从未被用户听过。在上传到Apple Music的内容中，AI生成的音乐占比在三分之一以上。Adam Mosseri, the head of Instagram, said late last year that it would soon “be more practical” to trace “real media than fake media.” His observation came not long after Meta, Instagram’s parent company, rolled out an all-A.I. video stream derided by many as a showcase for slop.Instagram负责人亚当·莫塞里在去年年底表示，追踪“真实媒体将比追踪虚假媒体更实际”。他的观察是在Meta（Instagram母公司）推出一个全AI视频流不久后发表的，该视频流被许多人嘲笑为垃圾内容的展示窗口。Many tech companies now trying to manage A.I. slop are also the ones enabling it. In July, Meta promised to support a European pact to make A.I.-generated content more identifiable via detection, labeling and tagging. That same month, the company also released an image generator that automatically opted in public Instagram accounts as a reference for the tool. (Meta removed the feature after three days of backlash.)许多现在试图管理AI垃圾内容的科技公司也正是助长它的那批公司。7月，Meta承诺支持一项欧洲协议，通过检测、标记和标注使AI生成的内容更容易识别。同月，该公司还发布了一款图像生成器，自动将公开的Instagram账户作为该工具的参照。（经过连续三天的舆论抵制，Meta移除了该功能。）“A.I. slop is a double-edged sword in that a lot of these platforms experienced upticks in engagement as a result of it,” said Paul Shovlin, an expert in A.I. at Ohio University. “But then, it got to a point where it became a liability because there’s so much flotsam and jetsam that it’s now hard for users to find the kind of quality that they’re looking for.”“AI垃圾内容是一把双刃剑，因为许多平台因此获得了参与度的提升，”俄亥俄大学AI专家保罗·肖夫林表示，“但后来它变成了一种负担，因为垃圾太多了，用户现在很难找到他们想要的那种质量的内容。”Google, for example, has gone all in on A.I., linking its Gemini model to its email, maps and travel services and search. More than 20 percent of the initial short videos on YouTube served to new users are A.I.-generated slop, according to an experiment conducted last year by Kapwing, an online video editor that is powered by A.I.例如，谷歌已全力投入AI，将其Gemini模型与电子邮件、地图和旅行服务以及搜索相连接。根据去年由在线视频编辑器Kapwing进行的一项实验，YouTube上为新用户推送的初始短视频中有超过20%是AI生成的垃圾内容。位于加利福尼亚州山景城的谷歌园区，该公司移除了5万个账号群组。 Kelsey McClellan for The New York TimesOne of the most popular YouTube slop channels, Kapwing estimated, earned $4.25 million annually. (It has since been suspended from a YouTube monetization program.) Neal Mohan, YouTube’s chief executive, wrote in a blog post that “managing A.I. slop” was a top priority this year.Kapwing估计，最受欢迎的YouTube垃圾内容频道之一每年赚取425万美元。（该频道此后已被暂停参与YouTube的创收计划。）YouTube首席执行官尼尔·莫汉在一篇博客文章中写道，“管理AI垃圾内容”是今年的首要任务。Recently, researchers at Google considered the challenge. Rather than evaluate videos individually, they looked at upload pace, repetitive templates and other patterns that could signal coordinated abusive behavior. They found 50,000 clusters of accounts, which were removed. Some digital media experts, however, worried that the approach could overreach and also penalize legitimate content creators who publish large volumes of content.最近，谷歌的研究人员考虑了这一挑战。他们没有逐个评估视频，而是着眼于上传速度、重复模板和其他可能表明协同滥用行为的模式。他们发现了5万个账号群组，并将其移除。然而，一些数字媒体专家担心这种方法可能矫枉过正，也会惩罚那些发布大量内容的合法创作者。“While academic research helps advance industry understanding, it doesn’t reflect the many layers of defense we use to enforce our policies,” YouTube said in a statement. “This includes robust safeguards to ensure well-intentioned creators using A.I. tools aren’t impacted.”“虽然学术研究有助于推动行业理解，但它并不能反映我们用于执行政策的许多层防御措施，”YouTube在一份声明中表示，“这包括强有力的保障措施，以确保使用AI工具的善意创作者不会受到影响。”Other companies are turning to a grab bag of strategies to address A.I. slop, including rallying their own patrons to help identify it. In October, Pinterest updated its system to allow users to limit how many A.I.-generated posts they see. TikTok said it was testing a feature in the United States that lets people dial up or down the amount of visible A.I. content. (The company also said it had automatically labeled more than three billion A.I.-generated videos.) TikTok removed more than 377,000 videos in the first three months of the year for violating its A.I. policies.其他公司正在转向各种策略来应对AI垃圾内容，包括动员自己的用户帮助识别。10月，Pinterest更新了系统，允许用户限制他们看到的AI生成帖子数量。TikTok表示正在美国测试一项功能，让用户可以自行调节看到的AI内容量。（该公司还表示已自动标记了超过30亿个AI生成的视频。）TikTok在今年前三个月删除了超过37.7万个违反其AI政策的视频。10月，Pinterest更新了系统，允许其虚拟展板用户限制他们看到的AI生成帖子的数量。 PinterestA.I. was used to create nearly half of the posts with more than 50 words on X, Elon Musk’s social media platform, according to Pangram, an A.I. detection start-up. To deal with “a tremendous amount of A.I. slop & reply spam,” the social media platform this year banned apps that effectively encourage bots to post by paying for content, wrote Nikita Bier, who was X’s head of product until this month.根据AI检测初创公司Pangram的数据，在埃隆·马斯克的社交媒体平台X上，字数超过50的帖子中有近一半是使用AI创建的。本月刚卸任X产品主管的尼基塔·比尔写道，为了应对“大量的AI垃圾内容和回复垃圾”，该社交媒体平台今年禁止了那些通过付费鼓励机器人发帖的应用。Chris Best, the chief executive of Substack, wrote last month that “it’s getting harder to tell what’s real on the internet,” and that “platforms that reward fakeness will create a race to the bottom.” The newsletter platform announced its own A.I. detection tool for users, powered by Pangram, which is designed to scan replies, comments and posts longer than 100 words and decide whether they were likely written by hand or with A.I. help.Substack首席执行官克里斯·贝斯特上个月说，“在互联网上越来越难分辨什么是真实的，”并且“奖励虚假内容的平台将导致逐底竞争”。这个通讯平台宣布为用户推出自己的AI检测工具，由Pangram提供支持，旨在扫描回复、评论和超过100字的帖子，并判断它们可能是人工撰写还是借助AI完成的。Some newsletter writers complained that the tool mistook posts written by humans as A.I.-generated, did not account for A.I.’s potential role in the creative process and put undue pressure on authors to defend their reputations against an opaque metric.一些通讯作者抱怨说，该工具将人类撰写的帖子误判为AI生成，没有考虑到AI在创作过程中的潜在作用，并给作者带来了不必要的压力，要求他们根据一个不透明的标准来证明自己的清白。This spring, Camille François and nearly two dozen other A.I. experts convened at a Slop Salon event in Menlo Park, Calif., where they discussed the difficulty of classifying, and thus managing, A.I. slop.今年春天，卡米尔·弗朗索瓦和二十多位AI专家在加利福尼亚州门洛帕克的一次“垃圾沙龙”活动中会面，讨论了分类并藉此管理AI垃圾内容的困难。“The hodgepodge of tactics is a symptom that we’re looking at a systemic problem with our information environment and not just a bad class of content called slop that we can just define and remove,” Dr. François, who studies harmful digital technologies at Columbia University, said in an interview. “A.I. is a creativity tool at the end of the day, and multifaceted, and platforms that handle this in a poor way will see people walk and go to other places.”“这种策略大杂烩表明，我们面临的是信息环境中的一个系统性问题，而不仅仅是一类可以简单定义和移除的名为垃圾内容的坏内容，”在哥伦比亚大学研究有害数字技术的弗朗索瓦接受采访时说，“AI归根结底是一种创造工具，而且具有多面性，那些处理不当的平台将会看到用户流失并转向其他地方。”Tiffany Hsu报道错误信息、虚假信息及其起源、动向和后果。她从事新闻行业已有20多年。翻译：纽约时报中文网点击查看本文英文版。获取更多RSS：https://feedx.net https://feedx.site