AI 学数据分析，先造一个假世界
训练 AI 做数据分析，最缺的不是算力，是「题」:真实数据涉及隐私和版权，不能拿来练；自己编数据又怕答案太假、模型学不到真本事。这篇的思路是反着来：先定义这个世界背后的规则(比如「这个城市的房价由地段、面积、学区决定」)，再按规则生成文件和任务。这样每个任务都有唯一正确答案，机器能自己批改，还能自动生成海量练习题。研究者用这套引擎造了 2135 道训练题和 100 道评测题，拿 10 个顶级模型一测，发现它们普遍会「算得快但想不深」——能算出数字，却总结不出洞察。用这些合成题训练出的 35B 模型，在多个数据分析基准上比原版模型涨了 7 到 20 分。它不是你明天能用上的工具，但指向一个趋势：AI 的下一个战场不是聊天，而是替人读表、找结论，而训练它的方式，正在从「喂真实数据」转向「造可控的假世界」。
📄 原文摘要(英文)
Data agents use code to process heterogeneous files and derive analytical insights. Scaling their training faces costly environment curation, privacy and intellectual-property restrictions, and difficulty verifying open-ended findings without known generative truth. Our core insight is to specify the world behind the files before generating the files themselves. We introduce Data World, a multi-agent engine that synthesizes environments, tasks, and rubrics from explicit mechanisms. Independent audits check target recoverability from observed files; rubrics support evaluation and RL rewards. Human preference-based curation for question realism selects 100 tasks across 50 held-out worlds and 16 domains for DataWorld-Bench. The engine also yields 2,135 training tasks from 1,037 disjoint worlds. Evaluation of ten frontier models shows that the benchmark remains challenging, particularly for insight discovery, with a consistent computation-insight gap. RL-zero training on the synthesized tasks produces DataWorld-35B-A3B from Qwen3.6-35B-A3B without trajectory distillation. Under matched evaluation, it improves over its base by 7.06 points on DataWorld-Bench, 14.28 on DAComp, 20.90 on DABstep-Hard, and 12.13 on JobBench, a broader occupational benchmark.