人工智能训练公司 micro1 将在未来 12 个月内斥资 10 亿美元，从各公司购买并获取运营数据的授权。这家位于旧金山的公司于周五公布了该计划。花旗银行和 Hercules Capital 将为其提供资金。
AI training company micro1 will spend $1bn over the next 12 months buying and licensing operational data from companies. The San Francisco company announced the plan on Friday. Citi and Hercules Capital are providing the capital.
这笔资金将通过 micro1 的“企业数据合作伙伴计划”进行运作。该公司利用经过脱敏处理的企业数据，构建能够反映真实业务运营的强化学习环境。在这些环境中，人工智能模型和智能体将学习处理工作流程、做出决策并完成复杂任务。
The money goes through micro1’s Company Data Partnerships programme. The company uses de-identified company data to build reinforcement learning environments that reflect real business operations. In them, AI models and agents learn to work through workflows, make decisions and complete complex tasks.
micro1 在公告中表示：“真实的业务运营涉及信息不完整、相互冲突的优先事项以及需要判断力的例外情况。”
“Real business operations involve incomplete information, competing priorities and exceptions that require judgment,” micro1 said in its announcement.
该公司称，这 10 亿美元将扩大其环境所覆盖的行业和工作流程范围。同时，该计划还能为企业开辟一条新的收入来源，使其能够利用多年来积累的运营知识创造收益。micro1 的采购内容该计划的页面列出了 micro1 所需的数据类型，包括标准作业程序、知识库、内部文件、客户关系管理（CRM）数据、项目历史记录以及质量保证流程。此外，该公司还希望获取团队决策记录以及对人工智能输出结果的人工反馈。
The company said the $1bn will widen the range of industries and workflows its environments cover. It said the initiative gives businesses a new revenue stream from the operational knowledge they have built over years. What micro1 pays for The programme’s page lists the data micro1 wants. That includes standard operating procedures, knowledge bases, internal documents, CRM data, project histories and quality assurance processes. It also wants records of how teams make decisions, and human feedback on AI outputs.
该公司表示，支付金额取决于数据的质量、独特性和价值。该页面列出了三种支付等级：符合条件的合作伙伴可获得超过 10 万美元；大型数据集或涉及多个团队的长期合作可获得超过 50 万美元；而高度独特且受版权保护的数据则可获得超过 100 万美元。
Payment depends on the quality, uniqueness and value of the data, the company said. The page lists three payment levels. A qualified partnership earns more than $100,000. Large datasets, or ongoing work across several teams, earn more than $500,000. Highly unique, proprietary data earns more than $1m.
企业仍保留其原始数据的所有权。该公司表示，会在“必要情况下”移除个人信息，也可能对部分数据集进行合成性改写，以削弱其与原始记录的直接关联。在 micro1 使用数据之前，企业可以审查数据样本。
Companies keep ownership of their underlying data. The company said it removes personal information “when relevant”. It may also rewrite some datasets synthetically to loosen their link to the original records. Companies can review samples before micro1 uses the data.
该计划面向拥有超过 30 名员工且具备完善文档体系的公司。该公司优先选择美国企业，其次是其他西方市场，且相关数据必须为英文内容。
The programme targets companies with more than 30 employees and established documentation. The company gives priority to US companies, then other Western markets, and the material must be in English.
Spirit航空公司记录竞标者 该公司自称是一家数据实验室，为人工智能研究公司和企业训练前沿模型及评估智能体提供帮助。它运营着三个内部实验室，其中一个专注于机器人技术。
Bidder for Spirit’s records The company calls itself a data lab that helps AI research companies and enterprises train frontier models and evaluate agents. It runs three in-house labs, one of them for robotics.
今年9月，micro1曾对Spirit航空公司的客户记录提出竞争性报价。此后，Spirit破产案中的法院指定隐私官员支持了谷歌以1000万美元收购该数据的交易。埃隆·马斯克的SpaceX也曾考虑从初创公司购买数据，用于训练其人工智能模型。
In September, micro1 made a competing bid for Spirit Airlines’ customer records. The court-appointed privacy official in Spirit’s bankruptcy has since backed Google’s $10m deal for that data. Elon Musk’s SpaceX has also weighed buying data from startups to train its AI models.