Inside the AI data boom: from labeling to RL environments
In this Silicon Valley 101 podcast episode, He Yunzhong, who works on post-training and evaluation research at Scale AI, and Sun Yiyou, a UC Berkeley postdoc on the Agents’ Last Exam (ALE) project, unpack the fast-growing but opaque market for AI training data. They note AfterQuery was valued at $300 million in its April A round and, per media reports, $3.2 billion by September, while Bespoke Labs announced $40 million in combined seed and A funding in July, and trace the shift from crowd labeling and pre-training toward rubrics, verifiable signals and RL environments bundling tasks, tools and verification. They also discuss procurement bottlenecks in vertical domains, the blurred line between benchmarks and data sales — including a claim that some data firms sold evaluation data — expert fabrication and contamination, and OpenAI's February decision to stop reporting SWE-bench Verified scores. Guest views are personal.
Data supply chains and benchmark credibility are becoming central bottlenecks for model progress, and this episode offers an on-the-ground view from both vendors and researchers.
随着AI模型能力提升，训练需求更复杂，一批为模型公司提供训练数据的新公司正在快速增长：AfterQuery今年4月宣布A轮融资时，估值为3亿美元，到了9月，据媒体报道，新一轮融资已经将其估值推至32亿美元。另一家提供训练数据与智能体环境的公司Bespoke Labs，也在今年7月宣布，种子轮与A轮融资合计达到4000万美元。
虽然估值水涨船高，AI数据行业对外界来说却极其不透明。这些公司究竟在卖什么？
过去提到数据标注，我们容易想到给图片打标签、给模型回答打分。但随着AI从回答问题走向执行任务，数据公司的交付物也在变化：从专家写出的评分标准，到包含任务、工具和验证机制的强化学习环境。它们需要找到懂行的人，获得真实的行业资料，再把专业知识与工作经验转化成模型能够学习的训练信号。
与此同时，新的AI评测层出不穷。榜单分数提高了，究竟代表哪些能力变强了？针对评测生产数据，什么时候能改善真实用户体验，什么时候又会变成单纯的刷分？对于每天训练模型的研究员来说，他们最需要的数据，到底是什么？
#Scale AI#Mercor#Surge AI#AfterQuery#Bespoke Labs#Agents’ Last Exam#Humanity’s Last Exam#SWE-bench Verified#OpenAI#Snorkel AI#University of California, Berkeley