本文提出Athanor系统,首次证明Claude Opus 5.5智能体团队可自主后训练Qwen3-4B-Base,使其性能接近或超过官方指令微调版本,为自主AI研究提供基础。
摘要:
递归自我改进(即AI系统构建更强的AI系统)是实现超级智能的一条有前景的路径。本文研究了一种当前可衡量的较弱形式:前沿智能体利用公开数据对小型原始基础模型进行后训练,以此作为其研究潜力的测试。我们构建了Athanor,这是一个面向真实GPU的自主后训练研究(AutoPostTrain)的开源智能体原生系统。我们为智能体提供完全的研究自主权、低摩擦的工程工具和高效的入门代码。智能体自行开展研究:它们读取数据集和评估结果、构建训练数据、编写和修改训练算法、调整超参数,并分析结果以决定下一步尝试的方向。它们在沙箱中工作,无法使用外部教师模型。Athanor的服务提供冻结评估、托管作业、去污染数据、持久研究记录以及智能体未知的保留测试集。在一场为期104小时的实验中,由四个Claude Opus 5.5智能体组成的团队(每个智能体配备8个GPU节点且未获得任何研究建议),在包含6个任务的开发套件上将Qwen3-4B-Base的性能从15.9提升至67.3,在保留测试集上从9.2提升至48.2,而官方Qwen3-4B-Instruct-2507的对应性能分别为66.5和47.9。据我们所知,这是首次证明前沿智能体团队可自主将原始基础模型后训练为跨领域通用模型,其平均性能与官方指令微调版本相当或更优。该实验的完整记录显示,智能体通过组合已知方法和彼此的模型达到这一水平:在公开推理数据上的监督微调(SFT)、采用某一智能体引入的rollout变化的强化学习(RL)、跨智能体的权重合并以及多轮自蒸馏。随着执行环节的解决,当前的智能体已具备完成该任务的工程和研究能力。我们希望Athanor能为扩大自主AI研究提供基础。
英文摘要:
Recursive self-improvement, in which AI systems build stronger AI systems, is one promising route to superintelligence. We study a weaker form that can be measured today: frontier agents post-training a small raw base model with public data, as a test of their research potential. We build Athanor, an open agent-native system for autonomous post-training research (AutoPostTrain) on real GPUs. We give the agents full research autonomy, low-friction engineering tools, and efficient starter code. The agents carry out the research themselves: they read datasets and evaluation results, build their training data, write and modify training algorithms, tune hyperparameters, and analyze the results to decide what to try next. They work in a sandbox and cannot use external teacher models. Athanor's services supply frozen evaluation, managed jobs, decontaminated data, a persistent research record, and a held-out suite the agents are unaware of. In a 104-hour campaign, a team of four Claude Opus 5.5 agents, each on an eight-GPU node and given no research advice, improves Qwen3-4B-Base from 15.9 to 67.3 on a six-task development suite and from 9.2 to 48.2 on a held-out suite, against 66.5 and 47.9 for the official Qwen3-4B-Instruct-2507. To our knowledge, this is the first demonstration that a team of frontier agents can autonomously post-train a raw base model into a cross-domain generalist model that matches or surpasses the official instruct version on average. The campaign's full record shows that the agents reach this level by recombining known methods and each other's models: SFT on public reasoning data, RL with a rollout change one agent introduced, weight merges across agents, and rounds of self-distillation. With execution handled, today's agents already have the engineering and research ability to conquer this task. We hope Athanor provides a foundation for scaling autonomous AI research.
6G Needs Agents: Toward Agentic AI-Native Networks for Autonomous Intelligence
Department of Computer and Network Engineering, United Arab Emirates University, UAE(计算机与网络工程系,阿联酋大学);Research Institute for Digital Future, Khalifa University, UAE(数字未来研究院,哈利法大学)
SysEvolve: An AI-native, safe, autonomous adversarial attack-defense co-evolutionary system
School of Computer Science, Peking University(北京大学计算机学院);Southeast University(东南大学);School of Computing, National University of Singapore(新加坡国立大学计算机学院)
SEAL: An Open, Auditable, and Fair Data Generation Framework for AI-Native 6G Networks
CONNECT Centre, Munster Technological University(芒斯特理工大学CONNECT中心);Centre Tecnològic de Telecomunicacions de Catalunya (CTTC/CERCA)(加泰罗尼亚电信技术中心);CONNECT Centre, School of Computer Science, University College Dublin(都柏林大学计算机科学学院CONNECT中心)
University of Rajshahi(拉杰沙希大学);Rajshahi University of Engineering and Technology(拉杰沙希工程技术大学);Universiti Kebangsaan Malaysia(马来西亚国民大学);Independent University, Bangladesh(孟加拉独立大学)
Owl-AuraID 1.0: An Intelligent System for Autonomous Scientific Instrumentation and Scientific Data Analysis
Shenzhen Loop Area Institute(深圳河套学院);The Chinese University of Hong Kong(香港中文大学);Dalian University of Technology(大连理工大学);Suzhou Laboratory(苏州实验室);The Hong Kong University of Science and Technology(香港科技大学)
Heterogeneity-Aware Belief Synchronization for Semantic Communication in AI-Native 6G Networks
School of Electrical Engineering and Computer Science (SEECS), National University of Sciences and Technology (NUST)(国家科学技术大学 电气工程与计算机科学学院);Kyung Hee University(庆熙大学);Munster Technological University(芒斯特理工大学)
Agentic Artificial Intelligence (AI): Architectures, Taxonomies, and Evaluation of Large Language Model Agents
University College of Engineering, Anna University(安娜大学工程学院);National Institute of Technology Tiruchirappalli(Tiruchirappalli 国家理工学院);School of Computing and Information Systems University of Melbourne(墨尔本大学计算机与信息系统学院)
Autonomous computational catalysis through an agentic research system
Beijing Key Laboratory of Artificial Intelligence for Advanced Chemical Engineering Materials, State Key Laboratory of Chemical Engineering and Low- Carbon Technology, Department of Chemical Engineering, Tsinghua University(北京先进化工材料人工智能重点实验室、化学工程与低碳技术国家重点实验室、清华大学化学工程系)
AINTMA: Agentic AI Architecture for Autonomous Test Management with Generative Intelligence, Secure Cloud Communication and Adaptive Quality Analytics
International Business Machines (IBM)(国际商业机器公司(IBM));Global Atlantic Financial(全球大西洋金融公司);Docusign(DocuSign公司);Salesforce Inc(Salesforce公司);The Home Depot(家得宝公司)
本文提出Claw AI Lab,一种自主研究平台,通过隐藏的提示到论文流程实现自动化研究,并提供交互式AI实验室。该平台允许用户通过一个提示创建完整的研究团队,支持自定义角色、协作流程、实时监控、 artifact检查和回滚/恢复控制。Claw-Code Harness连接本地代码库、数据集和检查点,提高实验执行、完成和结果完整性。在内部评估中,Claw AI Lab在想法新颖性、实验完整性和论文质量上被AI专家评委一致偏好。
Thousand-GPU Large-Scale Training and Optimization Recipe for AI-Native Cloud Embodied Intelligence Infrastructure
AI Infra Team at JDT(JDT人工智能基础设施团队);Tsinghua University(清华大学);Peking University(北京大学);Tianjin University(天津大学);Beihang University(北洋大学);University of Science and Technology of China(中国科学技术大学)