Kinesiska startupen Z.ai lanserar billigare modell med öppen källkod
中国人工智能初创公司Z.AI以前称为Zhipu，在上海启动了最新的开源模型，GLM-4.4。这种新的AI模型旨在通过以明显更低的价格提供高性能，以与OpenAI和DeepSeek等知名参与者竞争。 GLM-4.5的表现与Claude 3.5十四行诗和[…]中国初创公司Z.AI启动更便宜的模型，开源量首次出现在AI新闻中。
IEEE Transactions on Fuzzy Systems, Volume 33, Issue 7, July 2025
1) Optimizing Deep Neuro-Fuzzy Network for ECG Medical Big Data Through Integration of Multiscale FeaturesAuthor(s): Xin Wang, Jianhui Lv, Byung-Gyu Kim, Bidare Divakarachari Parameshachari, Keqin Li, Dongsheng Yang, Achyut ShankarPages: 2027 - 20372) FDformer: A Fuzzy Dynamic Transformer-Based高效工业时
On Monday, July 28, the Government of Guam honored U.S. Marine Corps Brigadier General (Brig. Gen.) and Delegate Vicente “Ben” T. Blaz, a Guam native and the first CHamoru Marine to achieve the rank of general officer in the U.S. Armed Forces, by renaming Guam Route 3 as “Brigadier General Ben Blaz
OpenAI, Google, and Anthropic Commit to ‘AI Watermarking’ to Combat Deepfakes
请注意，我们无权提供任何投资建议。此页面上的内容仅用于信息目的。 In a major move to tackle the growing threat of AI-generated misinformation, leading tech firms including OpenAI, Google, and Anthropic have agreed to implement digital “watermarking” techniques to identify content created by artificial… Continue reading OpenAI, Goog
zkSync’s ZK Token Stumbles 12% After Initial Hype Fades
请注意，我们无权提供任何投资建议。此页面上的内容仅用于信息目的。 Zksync备受期待的ZK代币在大肆宣传的首次亮相仅几周后，价格下跌了12％。该令牌，一旦由于其2层扩展承诺而引起的城镇的讨论，……在最初的炒作后，Zksync的ZKSK代币偶然的绊脚石在最初的炒作后，最初的炒作最初的炒作在经济观察中首次出现了，继续阅读ZKSYNC的ZK代币绊脚石12％。
Bathylepeta wadatsumi Chen, Tsuda & Ishitani, 2025DOI: doi.org/10.3897/zse.101.156207AbstractTrue limpets in the gastropod subclass Patellogastropoda are familiar members of shallow-water rocky environments but are much rarer in the deep, with just three families adapted to bathyal depths or more.其
Rajya Sabha adjourned for fourth day in row
Rajya Sabha由于反对抗议而连续第四天面临中断，导致多次休会。诉讼开始于乌吉瓦尔·尼卡姆（Ujjwal Nikam）的宣誓和告别演讲，向泰米尔纳德邦（Tamil Nadu）的六名退休成员发表讲话。但是，反对派成员破坏了问题小时，要求就各种问题进行讨论，包括贾格迪普·达赫赫（Jagdeep Dhankhar）董事长辞职。
Why a tech start-up wants to pump your faeces deep underground
从Microsoft提供资金的Deep的启动拱顶表示，将人类废物存储在地下，可以将二氧化碳置于大气中，而Forever Chemicals（如Forever Chemicals）之外的污染物
#475 – Demis Hassabis: Future of AI, Simulating Reality, Physics and Video Games
Demis Hassabis is the CEO of Google DeepMind and Nobel Prize winner for his groundbreaking work in protein structure prediction using AI.Thank you for listening ❤ Check out our sponsors: https://lexfridman.com/sponsors/ep475-scSee below for timestamps, transcript, and to give feedback, submit questi
A.I. May Be the Future, but First It Has to Study Ancient Roman History
Google DeepMind的软件模型将更精确的日期放在了一个重要的拉丁文本上，该文本归功于罗马皇帝，以演示其能力。
AI helps reconstruct damaged Latin inscriptions from the Roman Empire
Google DeepMind和历史学家创建了一个名为AENEAS的AI工具，该工具可以预测刻在古罗马帝国的石墙和石墙和陶器的拉丁铭文中。
Google DeepMind’s new AI can help historians understand ancient Latin inscriptions
Google DeepMind揭开了新的人工智能软件，可以帮助历史学家恢复古代拉丁版画背后的含义和背景。埃涅阿斯可以分析用长天气的石头写的单词，以说何时何地刻有刻画。它遵循Google以前的考古工具iThaca，该工具还使用深度学习来重建和上下文化古代文本，…
DeepMind and OpenAI claim gold in International Mathematical Olympiad
在享有声望的年轻数学家竞争中，两种AI模型首次达到了金牌标准 - 他们的开发人员声称这些AI很快就会解决棘手的科学问题
'What Do All Those PhDs Do?' - Bessent Calls For 'Fundamental Reset' Of Financial Regulations
'所有这些博士都做什么？” - Bessent Calls For 'Fundamental Reset' Of Financial RegulationsTreasury Secretary Scott Bessent on Monday called for a “fundamental reset” of financial regulations to ensure they are aligned with the nation’s domestic and international priorities.Speaking at the Federal Reserve Capita
"No Spare Capacity": Watchdog Warns Largest US Grid Is Maxed Out Amid Data Center Buildout
"No Spare Capacity": Watchdog Warns Largest US Grid Is Maxed Out Amid Data Center BuildoutAmerica's largest power grid has issued multiple 'Maximum Generation' and 'Load Management' alerts this summer, as summer heat pushes power demand to the brink with air conditioners running at full blast across
DeepMind’s Quest for Self-Improving Table Tennis Agents
几乎没有一天，没有令人印象深刻的新机器人平台从全球学术实验室和商业初创公司出现。尤其是人形机器人看起来越来越有能力帮助我们在工厂，最终在家庭和医院中。但是，要使这些机器真正有用，他们需要精致的“大脑”来控制其机器人身体。传统上，编程机器人涉及专家花费无数小时精心编写复杂行为，并详尽地调整参数，例如控制器的增益或运动规划权重，以实现所需的性能。尽管机器学习（ML）技术有希望，但需要学习新的复杂行为的机器人仍然需要大量的人类监督和重新设计。在Google DeepMind，我们问自己：我们如何使机器人能够更加整体，连续地学习和适应，从而减少了每项重大改进或新技能的专家干预瓶装的瓶颈？这个问题是我们
组织正在采用大型语言模型（LLM），例如DeepSeek R1，以改变业务流程，增强客户体验并以前所未有的速度推动创新。但是，独立的LLM具有关键的局限性，例如幻觉，过时的知识和无法获得专有数据的访问。检索增强发电（RAG）通过将语义搜索与生成AI相结合，[…]
