腾讯混元发布 WorkForge:为长周期工作智能体构建可验证训练环境
腾讯混元发布 WorkForge 框架,称可从真实世界资源出发,为长周期工作智能体(long-horizon work agents)构建可验证的训练环境。该团队介绍,WorkForge 以专家工作流为起点,搭建真实工作空间、抽取可核查事实,并据此派生任务,同时配套程序化与语义两类验证器,均以可观测证据为依据。腾讯混元称该框架已扩展至 40 个专业领域、60 种文件类型、共 1.67 万个环境,并称随着数据规模与交互长度提升,在长周期工作基准上取得稳定增益;帖子附有 arXiv 论文链接。
腾讯将大规模可验证环境视为训练能完成长链条多步工作任务的智能体的路径,但所提到的基准提升目前仅来自该团队自述。
📢 Introducing WorkForge: a scalable framework that constructs verifiable training environments for long-horizon work agents from real-world resources. Starting from expert workflows, it builds authentic workspaces, extracts checkable facts, and derives tasks plus dual (programmatic + semantic) verifiers grounded in observable evidence. 🚀 We scale to 16.7K environments across 40 professional domains covering 60 file types, enabling consistent gains on long-horizon work benchmarks as data volume and interaction horizon increase. 📚Paper: https://arxiv.org/pdf/2610.04906 @zhang_leo54389 @ZoeyyyZhououou @Be1ong1 #Tencent Hy#AIAgents #WorkAgents #AgentTraining