01政策相关
3Claude 文本水印机制如何运作
28 天前未来 Claude 模型生成的文本将包含水印,用于判断文本由 Claude 撰写的可能性,这是 Anthropic 为遵守欧盟《AI 法案》而实施的变更。该方法基于 Google DeepMind 的 SynthID-Text 技术,对输出质量、创造力和可读性无实际影响,读者无法区分水印文本与普通文本,且不增加额外 token 或成本。
OpenAI and Anthropic in price war as Chinese AI rivals gain ground
28 天前Cursor 正式被 SpaceX 收购
28 天前Cursor 已被 SpaceX 正式收购,完成自 4 月启动的收购流程。合并后 Cursor 将获得全球最大 GPU 集群,以构建更强且运行成本更低的模型,从而以更低价格向客户提供更强大的模型。本周三发布的 Grok 4.6 是双方合作成果的早期体现。
02行业前沿动态
5Gemini 3.7 Flash 全面上线 Pro 与 Ultra 用户
28 天前Gemini 3.7 Flash 现已向 Gemini 聊天中的 Pro 和 Ultra 用户开放。该模型更新提升了多步骤任务的推理与准确性,如智能整合数十个文件和邮件为一份主文档。同时,Gemini Spark 也已运行于 3.7 Flash,通过改进对 Google Workspace 应用的工具调用,让个人 AI 智能体更精准。
通义千问开源 Qwen3.8 系列模型
28 天前通义千问兑现承诺,开源 Qwen3.8 系列模型。其中 Qwen3.8-27B 为原生多模态稠密模型,仅 27B 参数即全面超越 Qwen3.7-Plus,原生支持 262K 上下文,可通过 YaRN 扩展至 1M tokens,采用 Apache 2.0 许可。Max 级 Qwen3.8-2.4T-A95B 的开放权重也已同步发布。
dots3-note Preview 开源:280B 参数轻量模型,主打长程智能体与多模态推理
28 天前小红书技术开源 dots3-note Preview,这是 dots3 系列最轻量模型,总参数 280B、激活参数 16B,支持 512K 上下文及文本、视觉、语音多模态理解,并针对复杂推理和长程 Agent 任务优化。
GLM-5.3 发布:编程能力开源第一,并涌现网络安全能力
29 天前智谱发布GLM-5.3,基于与GLM-5.2相同的基座,通过极致的后训练Scaling提升智能上界,编程能力较前代提升50%,在Terminal Bench 3.0等公开基准中取得开源第一,并接近Claude Fable 5。模型在白盒代码审查等安全任务中表现持平Mythos 5,在CyberGym测试中得分84.5%。模型权重将在两周后开源,即日起上线ZCode、AutoClaw等工具。
DeepSeek V4 Pro 登陆硅基流动,1M 上下文
29 天前DeepSeek-V4-Pro-0813 正式上线硅基流动 SiliconFlow,提供 Day-0 支持,具备 1M 上下文窗口及低/高/最大三档推理强度,更侧重编码、工具调用与智能体工作流,仍保持 MIT 开源协议。定价为输入 $1.32/M、输出 $3.96/M、缓存命中 $0.44/M。同系列 DeepSeek-V4-Flash-0731 则面向追求速度与成本效益的日常生产场景。
04论文研究
8The Affordance is the Message: Creative Media as Complex Systems
29 天前arXiv:2608.12349v1 Announce Type: new Abstract: The affordances of a creative medium strongly condition the creative artefacts the medium will produce. In this work, we present a formalisation of computational creativity…
Visibility Asymmetry: How Vendor Attention Shapes Which EdTech Breakdowns Become Product-Visible
29 天前arXiv:2608.12353v1 Announce Type: new Abstract: Infrastructure scholarship in CSCW often treats breakdown as the moment when infrastructures become visible. However, in vendor-managed sociotechnical systems, not all brea…
Humans are Missing from AI Coding Agent Research
29 天前arXiv:2608.12355v1 Announce Type: new Abstract: Recent progress in AI coding agent research has led to rapid improvements in agents' ability to autonomously perform complex software engineering tasks, from editing large …
DrawTalking It Out: Creativity-Support Research as Creative Process Itself
29 天前arXiv:2608.12357v1 Announce Type: new Abstract: I think the creative process in conducting open-ended creativity-adjacent research is itself part of the creative process (including ideation, pivots, and tangents excluded…
Interaction Readiness: A Framework for Building and Evaluating AI Agents in Human Roles
29 天前arXiv:2608.12358v1 Announce Type: new Abstract: Product and engineering teams building role-bearing AI agents face an evaluation gap: an agent can produce accurate, safe, and fluent content while still failing the behavi…
What Do We Mean When We Talk About Infographics?
29 天前arXiv:2608.12370v1 Announce Type: new Abstract: There has been limited clarity and consistency regarding what the term infographics, or information graphics, refers to in visualization research and practice. In particula…
Transforming Interactions in Thesis Supervision: An Expos\'e-First Workflow in Higher Education
29 天前arXiv:2608.12546v1 Announce Type: new Abstract: At the studied research institute, one professorship oversees approximately 20 theses per semester, while day-to-day supervision is distributed among doctoral and postdocto…
Analysis of Motor Signatures of Social Adaptation in Autism for Efficient Human-Centric Systems
29 天前arXiv:2608.12548v1 Announce Type: new Abstract: Dance imitation integrates motor planning, sensorimotor integration, and social cognition, offering a sensitive framework to characterize motor behavior in autism. In this …
05技巧与观点
2蚂蚁百灵与 ASystem 团队打通单机 Agentic RL 后训练闭环
29 天前蚂蚁百灵与 ASystem 团队合作,用 Ling-3.0-tiny 和 AReno 在 DGX Spark 上跑通单机 Agentic RL 后训练闭环。以井字棋为最小验证任务,用 GSPO 算法训练 400 步后,rollout/rewards_mean 从约 -0.5 升至 0.4,response_len 降至约 850 tokens,工具调用与动作选择趋于稳定。
2026年夏季开源模型生态观察:中国前沿模型规模领先,AMD与NVIDIA主导发布量
29 天前2026年1至8月,Hugging Face公开模型仓库从243万增至296万,但85.6%的模型下载量不足200次,1.5%的仓库占据99.2%下载量。中国实验室月度最大开源模型参数规模在754B至2.78万亿之间,美国实验室七个月中五个月低于130B。AMD与NVIDIA各发布超200个新模型仓库,成为发布开源模型最多的机构。
