1 、 国务院发布深入实施 " 人工智能 +" 行动意见。意见明确，到 2027 年，率先实现人工智能与 6 大重点领域广泛深度融合，新一代智能终端、智能体 等应用普及率超 70% ，智能经济核心产业规模快速增长； 到 2030 年，我国人工智能全面赋能高质量发展，新一代智能终端、智能体等应用普及率超 90% ，智能经济成为我国经济发展的重要增长极。 2 、 国金证券调整融资保证金比例的消息引起市场关注。据悉，国金证券确有发布关于调整标的证券融资保证金比例的公告，但目前并无全行业保证金调 整的通知。 国金证券公告将于 8 月 27 日起，对除北交所以外的标的证券的融资保证金比例做出调整，新开立的除北交所以外的标的证券的融资合约适用 的融资保证金比例调整为 100% 。华创证券非银徐康团队认为，国金证券的调整更多是基于公司近期经营杠杆的考虑，无需过度解读。 1 、 国务院总理李强主持国务院第十五次专题学习时强调，要主动扩大优质服务进口，以高水平开放促进我国服务业高质量发展； 要推动服务贸易制度 型开放，加强与国际高标准经贸规则对接，进一步压减跨境服务贸易负面清单，有序放宽服务领域市场准入；加快建设全国统一大市场 ...

OpenAI is making changes to its popular chatbot following a lawsuit alleging that a teenager who died by suicide this spring relied on ChatGPT as a coach https://t.co/qOK0UxJZ8o ...

On this episode of Elon, Inc. we talk about Elon Musk’s companies suing Apple and OpenAI and whether Musk made up with rival Mark Zuckerberg https://t.co/ZoSxRkWZcD ...

OpenAI最新模型有什么特别之处？ 2025年8月，OpenAI推出最新版AI模型"GPT-5"，首席执行官萨姆·奥尔特曼称，这是"世界上最好的模 型"。"GPT-5"上线后，众多用户反馈，它虽在专业回答上有提升，但情感交互温度、个性及想象力不及 GPT-4o。 吴小莉：OpenAI的GPT-5出来后，大家抱以很大的期待，到底GPT-5有什么样的特殊性？ 杨红霞：大模型的整体能力其实是三个维度，第一层是聊天和记忆，这个其实不难做；第二层的能力是 推理和规划，就相对比较有挑战了；理论上GPT-5应该是它的第三层维度能力，比较特别，解决的是多 模态，但实话讲，多模态其实是比较难解决的，原因是语言经过了人类几千年的锤炼，不管是中国人、 美国人还是阿拉伯人也好；不管有人说话冗长一点，还是简短一点，但是信息的含量是相对比较对齐 的。 但是多模态这件事情不是这样子的，举个简单例子，哪怕是同样尺寸的图片，比方说都是256mm乘以 256mm的尺寸，但是你放大看可能就是一个茶杯、一个标签就足够了，缩小看可能是整个香港理工大 学，甚至整个香港，有很多的细节，需要很长的一段话。但是互联网上这么多的数据，怎么精准地把这 张图片和 ...

I fully understand that, you know, not only has this been sort of reported by some others and talked about by you, me and many others on our very desk, including, you know, some who some of our guests who have come on, you know, the the street has bandied about about this perplexity idea. Do we do we take in any way the continued reporting that what was a little bit of smoke uh seems to be gathering into a potential fire. Not well not really because these reports are also saying these discussions are intern ...

In Brief Meta plans to launch a super PAC to support California candidates favoring a light-touch approach to AI regulation, Politico reports. The news comes as other Silicon Valley behemoths, like Andreessen Horowitz and OpenAI’s Greg Brockman, pledge $100 million for a new pro-AI super PAC. Meta will pour tens of millions into its new group, dubbed Mobilizing Economic Transformation Across California, according to Politico. Brian Rice, Meta’s VP of public policy and head of the new PAC, has argued that Sa ...

Billionaire entrepreneur Elon Musk’s artificial intelligence startup xAI sued Apple and ChatGPT maker OpenAI in U.S. federal court in Texas on Monday, accusing them of illegally conspiring to thwart competition for artificial intelligence. Apple and OpenAI have "locked up markets to maintain their monopolies and prevent innovators like X and xAI from competing," the lawsuit said.Apple has partnered with OpenAI to integrate ChatGPT into its operating system for iPhones, iPads and Macs. Musk has been sharply ...

Two of the biggest private companies in the world, plus one of OpenAI's biggest competitors. You can invest in all three -- but is the price worth it?Everything and a bag of chips. That's what you get when you invest in Cathie Wood's Ark Venture Fund (ARKVX 3.08%), a mixed basket of 60-plus private and publicly traded stocks. It's a rare opportunity to buy shares of some of the most promising private companies in the market. Among them are SpaceX, OpenAI, and Anthropic.As of this writing, these privately ow ...

AI models are growing in terms of size and complexity and this is creating even more demands on the AI and data center infrastructure. In order to meet those requirements, we need to innovate at every level. We've talked a lot about all the enhancements we made in our hardware platforms, but we're also doing incredible innovations on the algorithmic side.We're introducing a new FP4 format called MVFP4. While we still support MXFP4, which is the open standard, we've invented MVFP4 to drive more performance a ...

一、 英伟达最新推出Jet-Nemotron小模型系列（2B/4B） 1. Jet-Nemotron是英伟达最新推出的小模型系列，由全华人团队打造，提出后神经架构搜索(PostNAS)与新型线性 注意力模块JetBlock； 2. 模型在数学、代码、常识、检索和长上下文等维度表现突出，性能超越Qwen3、Gemma3、Llama3.2等主流开源 全注意力语言模型； 3. 在H100 GPU上推理吞吐量最高提升53.6倍，长上下文场景下的优势特别明显，是英伟达在小模型领域的重要布 局。 https://mp.weixin.qq.com/s/8ZbWGnogg40sHknVBWHH1Q 二、 面壁多模态新旗舰MiniCPM-V 4.5：8B 性能超越 72B 生成式AI 1. 面壁小钢炮MiniCPM-V 4.5成为首个具备"高刷"视频理解能力的多模态模型，8B参数量却超越Qwen2.5-VL 72B 模型； 2. 该模型在MotionBench、FavorBench榜单达到同尺寸SOTA，最大可接收6倍视频帧数量，达到96倍视觉压缩 率； 3. 采用3D-Resampler高密度视频压缩、统一OCR和知识推理学 ...