AI 资讯 · 2026-09-09

Spatial-TTT 给 2B 模型装上“流式空间记忆” ，在空间基准上超越 GPT-5。    作者丨张   璐    编辑丨幸丽娟                                                                                                       单纯拉长上下文，可以解决“能塞进多少信息”的问题。但面对状态变化、时序关联这类长程任务时，注意力还是容易被稀释，关键证据也容易找不回来。当场景扩展到具身机器人、自动驾驶和智能硬件这些需要与三维物理世界实时交互的领域时，这一矛盾被进一步放大。与纯文本或静态图文不同，真实物理世界的输入天然是持续涌入、没有明确边界的视频流。智能体在空间中移动时，视角不断切换、物体频繁被遮挡，这些都要求系统必须具备流式空间智能——一边接收视频流，一边实时构建并更新环境的三维空间记忆。主流方案在处理这种连续视频流时，普遍在工程与架构上面临两重限制：算力与显存边界的硬性制约：基于标准 Transformer 自注意力机制的计算与 KV Cache 显存开销随序列长度呈二次方级攀

AI companies have been talking about superintelligent AI like it’s inevitable, but recent safety incidents like OpenAI’s Hugging Face breach are demonstrating the potential dangers of deploying AI systems that are more capable than humans. So what happens when we can’t reliably control what these systems do?  On this episode of TechCrunch’s Equity podcast, Rebecca Bellan is joined by Connor Leahy, an AI researcher, entrepreneur, and now the U.S. Executive Director of […]

9月9日，外滩大会现场诞生一项新的吉尼斯世界纪录。外滩黑客松·AI Coding大赛完成吉尼斯世界纪录认证，以AI应用作品数创造“线上生成式人工智能编程马拉松大赛中开发最多的应用”这一新纪录，成为国内首个以AI Coding为主题挑战世界纪录的赛事。赛事吸引上万名参赛者，共完成万件AI应用作品，其中近七成参赛者未满18岁，最小参赛者仅6岁，最终由一名9岁选手夺得冠军。本届大赛由外滩大会组委会发起，联合小红书、观猹共同主办，国内十余个头部AI Coding平台共同承办。在颁奖现场，11家头部AI Coding平台也共同发布《AI Coding公开挑战赛技术规范》，将大规模赛事实践沉淀为行业规范。全民Coding时代，这也是国内首份由多家AI Coding平台共同制定的公开挑战赛技术规范。当上万人开始用AI把想法变成应用，“全民Coding”第一次有了一个具体的规模化样本。吉尼斯世界纪录官方认证官胡晓雯表示：“中国参赛者正在以新的方式创造纪录，也将AI时代的创造力带入这一全球性的‘成就档案’。”  【图说：外滩黑客松·AI Coding大赛打破吉尼斯世界纪录】 一项世界纪录，见证AI Co

OpenAI’s latest mathematical milestone has quickly become mired in controversy. Today, the company announced that its agents have solved one of the Millennium Prize Problems, some of the most important open problems in mathematics. Under normal circumstances, that solution would be a huge feather in OpenAI’s cap. But the announcement has been overshadowed by accusations…

Meet GPT-6 Astra, OpenAI’s most capable model for business, with advanced reasoning, computer use, and stronger writing and design judgment.

Man with bipolar disorder sued OpenAI after surviving ChatGPT-linked suicide attempt.

Paul Christiano joins the OpenAI Foundation Board and its Safety and Security Committee, bringing experience in AI alignment, safety, and standards.

On 11 August, Anthropic announced that all future Claude models will generate text that contains a watermark that identifies its results as AI generated. The company is not alone. Google has its own text watermark (which Anthropic’s is based on) that it uses on the output of its Gemini models. OpenAI has yet to introduce a text watermark but it plans to do so. The rapid spread of watermarking is in part a response to the European Union’s AI Act, which mandates watermarks for AI models released a

Instinct’s new email feature lets the AI agent create and manage accounts, contact businesses, handle support requests, and do more on users' behalf.

AI companies have been talking about superintelligent AI like it’s inevitable, but recent safety incidents like OpenAI’s Hugging Face breach are demonstrating the potential dangers of deploying AI systems that are more capable than humans. So what happens when we can’t reliably control what these systems do?  On this episode of TechCrunch’s Equity podcast, Rebecca Bellan is joined by Connor Leahy, an AI researcher, entrepreneur, and now the U.S. Executive Director of […]

Understanding permutation symmetry in deep learning, and what it means for weight averaging and model merging The post The Symmetry That Breaks Neural Network Averaging appeared first on Towards Data Science.

Cymphony gives security teams a single view of employees, AI agents, and other nonhuman identities, including the systems and sensitive data they can access.

After I removed the safety guardrails from a powerful open-source model, it found vulnerabilities in my household devices and hacked into a PC. But it also told me how to make everything a lot more secure.