微软 Echoverse 用合成环境训计算机操作智能体，9B 从 36.5% 涨到 67.1%，还开源四个基准。
微软提出 Echoverse，把规格说明编译成有状态应用，并用应用自身数据库给任务评分。系统运行共进化循环，把每条已评分的执行轨迹读两遍：一遍修复环境、任务和验证器，一遍作为模型训练信号。浅层合成环境把线上站点准确率从 80.0 降到 75.0，深层环境把它从 80.0 提到 85.0，另一个领域从 48.0 提到 65.0。修复单个环境后，训练出的模型从 16.2% 升到 38.5%；用十二个环境训练，9B 模型在十四个评测分项上从 36.5% 升到 67.1%，与教它的更大前沿模型差距在 14 个百分点以内。团队发布四个环境作为基准，包含应用、种子数据和基于应用数据库的评分器。
New research from Microsoft. This one is on training computer-use agents at scale. Recent pipeline...
New research from Microsoft. This one is on training computer-use agents at scale. Recent pipelines generate synthetic environments in bulk, which moved the bottleneck from how many exist to what is inside each one. Echoverse compiles specifications into stateful applications whose tasks are graded against the application's own database, then runs a co-evolution loop that reads every graded rollout twice. Once as repairs to the environment, its tasks and its verifier. Once as training signal for the model. On the same domains, shallow environments pushed live-site accuracy below the base model, from 80.0 down to 75.0. Deep ones raised it, 80.0 to 85.0 and 48.0 to 65.0. Repairing a single environment lifted the model trained on it from 16.2% to 38.5%. Across twelve environments, a 9B model went from 36.5% to 67.1% on fourteen evaluation splits, within fourteen points of the much larger frontier model that taught it. They release four environments as a benchmark with applications, seed data and grounded graders. Paper: arxiv.org/abs/2607.28074 Track trending AI papers in our academy: academy.dair.ai 💬 4 🔄 1 ❤️ 11 👀 1983 📊 6 ⚡