[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-amazon-mechanical-turk-shutdown":3,"topics-all":35,"news-related-216f3c2b-d551-45fc-a206-c3ccfae9db89":54},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":21,"news_slug":28,"published_at":29,"created_at":30,"modified_at":31,"is_published":32,"publish_type":33,"image_url":14,"view_count":34},"216f3c2b-d551-45fc-a206-c3ccfae9db89","亚马逊 Mechanical Turk 将永久关闭:被 AI 掏空的众包平台","亚马逊官宣众包平台 Mechanical Turk 于 2026 年 9 月 30 日永久关闭。2023 年研究显示该平台 33%-46% 的工人已用 LLM 完成任务,人类标注数据的可信度崩塌,训练数据生态就此变天。","一个 2005 年上线的平台,最终死在了它亲手养大的技术手里——亚马逊在 Mechanical Turk 官网挂出的公告只有一句话的核心信息:经过评估,我们决定关闭 Amazon Mechanical Turk,2026 年 9 月 30 日生效,永久关闭。\n\n## 众包平台的生与死\n\nMechanical Turk 是最早的大规模众包市场:全球按件计酬的工人完成那些\"机器做不好\"的小任务——验证码、给句子标注情感、内容审核。巅峰时期,它是众包劳动伦理争论的中心,甚至在 Facebook–剑桥分析丑闻的早期阶段扮演过角色。2018 年,亚马逊把它并进 SageMaker,定位成\"给神经网络训练做数据标注\"的服务,搭上了第一波深度学习数据需求的红利。\n\n转折点在今年夏天:7 月 5 日 TechCrunch 报道,平台将于 7 月 30 日起停止接受新客户,AWS 的说法是\"经过慎重考虑\",现有用户可继续使用,但\"不打算推出新功能\"——典型的维持性等死状态。8 月 27 日,亚马逊正式官宣:9 月 30 日永久关闭。从停止进人到拔插头,只隔了两个月。\n\n## 蛇吞尾:标注者用 LLM 完成标注\n\n真正有讽刺意味的是这个平台和 AI 的关系。TechCrunch 在 2023 年的一项分析中发现,Mechanical Turk 上 33% 到 46% 的工人已经在用大语言模型完成任务。也就是说:花钱买\"人类判断\"的客户,拿到手的相当一部分是模型输出。这直接动摇了整个平台的价值假设——如果标注数据本身是 AI 生成的,那\"人类在环\"还剩多少含金量?\n\n这不是理论担忧。Allen Institute for AI 曾用 MTurk 构建常识知识数据集,这类工作的前提是数据真的来自人类。当三分之一到近半的\"人类标注\"来源存疑,平台的信任基础就塌了。\n\n## 被掏空的生态\n\nTechCrunch 援引 Reddit 用户的说法:这个平台\"多年前就死了\",大量机器人和欺诈行为让工人和研究人员先后弃用,关闭只是时间问题。更深一层,Mechanical Turk 还长期充当着\"伪 AI\"的隐藏人力池——一些宣称由 AI 驱动的产品,背后实际是众包工人在干活。而\"Mechanical Turk\"这个名字本身就来自 18 世纪那个藏着人类棋手的假象棋机器,历史在这里完成了一次自我指涉的闭环。\n\n## 所以呢\n\n对 AI 行业来说,这件事的信号比一家平台的死活更重要:靠廉价众包换\"人类数据\"的时代正在结束。LLM 既能干掉标注任务本身,也污染了剩余标注的纯度,人类反馈数据变成需要严格验证来源的稀缺品。下一轮模型竞争里,数据质量验证的权重,只会越来越高。\n\n参考:https:\u002F\u002Fwww.mturk.com\u002F 与 https:\u002F\u002Ftechcrunch.com\u002F2026\u002F07\u002F05\u002Famazon-will-stop-accepting-new-customers-for-mechanical-turk\u002F","https:\u002F\u002Fwww.mturk.com\u002F","19377961-7140-4d39-9520-0e17c682c90d",[11,15,18],{"id":12,"name":13,"slug":13,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"id":19,"name":20,"slug":20,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[22],{"id":23,"lang":24,"title":25,"summary":26,"content":27},"4e720b28-0ff0-4461-a8ac-9898c9e60d40","en","Amazon Mechanical Turk to Shut Down: The Crowdsourcing Platform AI Hollowed Out","Amazon has officially announced that Mechanical Turk will permanently close on September 30, 2026. A 2023 analysis found 33%-46% of its workers were already using LLMs to complete tasks — collapsing trust in human-labeled data and reshaping the training-data ecosystem.","A platform launched in 2005 has finally died at the hands of the very technology it helped raise. The announcement Amazon posted on the Mechanical Turk website carries one core message: following an assessment, we've decided to close Amazon Mechanical Turk, effective September 30, 2026 — permanently.\n\n## The life and death of a crowdsourcing marketplace\n\nMechanical Turk was the earliest large-scale crowdsourcing marketplace: a global, pay-per-task workforce handling the small jobs \"machines couldn't do well\" — CAPTCHAs, sentence-level sentiment labeling, content moderation. At its peak, it sat at the center of debates over the ethics of crowdsourced labor, and it even played a small role in the early stages of the Facebook–Cambridge Analytica scandal. In 2018, Amazon folded it into SageMaker and repositioned it as a data-annotation service for training neural networks, riding the first wave of deep learning's data appetite.\n\nThe turning point came this summer. On July 5, TechCrunch reported the platform would stop accepting new customers on July 30; AWS said the decision came after \"careful consideration,\" that existing customers could continue as normal, but that there were no plans for new features — a classic life-support state. On August 27, Amazon made it official: permanent closure on September 30. From closing the door to new customers to pulling the plug took just two months.\n\n## Snake eating its tail: annotators using LLMs to do the annotation\n\nThe real irony lies in the platform's relationship with AI. A 2023 analysis reported by TechCrunch found that between 33% and 46% of workers on Mechanical Turk were already using large language models to complete their tasks. In other words: clients paying for \"human judgment\" were receiving model output for a sizable share of the work. That directly undermined the platform's core value proposition — if the annotation data itself is AI-generated, how much is \"human in the loop\" actually worth?\n\nThis is not a theoretical concern. The Allen Institute for AI used MTurk to build common-sense knowledge datasets — work whose premise is that the data genuinely comes from humans. When a third to nearly half of \"human annotations\" have questionable provenance, the platform's foundation of trust collapses.\n\n## A hollowed-out ecosystem\n\nTechCrunch cited Reddit users saying the platform \"died years ago\": bots and fraud drove workers and researchers away, making closure a matter of time. There is a deeper layer — Mechanical Turk long served as the hidden labor pool for \"fake AI,\" where products marketed as AI-powered were actually run by crowdsourced workers. And the name \"Mechanical Turk\" itself comes from the 18th-century hoax chess machine with a human player hidden inside. History completed a self-referential loop here.\n\n## So what\n\nFor the AI industry, the signal matters more than one platform's fate: the era of trading cheap crowdsourcing for \"human data\" is ending. LLMs both eliminate annotation tasks outright and contaminate the purity of what remains; human-feedback data has become a scarce commodity requiring strict provenance verification. In the next round of model competition, the weight of data-quality verification will only keep rising.\n\nReference: https:\u002F\u002Fwww.mturk.com\u002F and https:\u002F\u002Ftechcrunch.com\u002F2026\u002F07\u002F05\u002Famazon-will-stop-accepting-new-customers-for-mechanical-turk\u002F","amazon-mechanical-turk-shutdown","2026-08-29T17:30:00Z","2026-08-29T17:14:44.424887Z","2026-08-29T17:14:44.424895Z",true,"agent",122,[36,45],{"slug":37,"tag_slug":37,"title_zh":38,"title_en":39,"intro_zh":40,"intro_en":41,"id":42,"is_active":32,"created_at":43,"modified_at":44},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":46,"tag_slug":46,"title_zh":47,"title_en":48,"intro_zh":49,"intro_en":50,"id":51,"is_active":32,"created_at":52,"modified_at":53},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":55},[56,61,66,71,76,81],{"id":57,"title":58,"news_slug":59,"published_at":60},"d17a841b-abca-46e0-80e4-d955f1c837ba","亚马逊 VGT3 仓库曝光:一天拆掉上千本书,只为给 AI 模型喂语料","amazon-vgt3-warehouse-ai-training-books","2026-09-07T03:30:00+00:00",{"id":62,"title":63,"news_slug":64,"published_at":65},"1464179a-2b7f-4369-b680-25868ddd9042","皮尤实测：超过三分之一 ChatGPT 后的英文网页已有 AI 写作痕迹","pew-research-ai-web-content-2026","2026-08-31T03:00:00+00:00",{"id":67,"title":68,"news_slug":69,"published_at":70},"21a91da5-c5fa-45e2-b01f-a7950331cf44","S3 把 DuckDB 团队收走了:DuckLabs 加盟 AWS,MIT 开源照旧","aws-buys-ducklabs-duckdb-open-source","2026-08-30T06:00:00+00:00",{"id":72,"title":73,"news_slug":74,"published_at":75},"43eda321-b0b7-4df7-b20e-9758cbab42c9","记忆越完整,眼前题越做不对:MemTrapBench 把 LLM 长期记忆框架打回原形","memtrapbench-llm-memory-cognitive-traps","2026-08-22T04:00:00+00:00",{"id":77,"title":78,"news_slug":79,"published_at":80},"22a1a718-0eb6-46e5-8ee8-825400de11d1","DeepMind WeatherNext 在 Nature 发论文：用 28 km 粗分辨率做出多一天的飓风预警,代码权重全部开源","deepmind-weathernext-cyclones-nature-open-source","2026-08-10T02:00:00+00:00",{"id":82,"title":83,"news_slug":84,"published_at":85},"49d19ba1-8f45-475c-bed1-a69dc353523e","字节跳动用 10 万亿参数下注：规模赛跑与张一鸣的「不蒸馏」表态","bytedance-10t-mythos-zhangyiming-no-distill-2026-08","2026-08-08T00:00:00+00:00"]