NAll news

All news

Tuesday792 items

  1. China.com Technology
    AI score42

    Gongji Technology Completes Funding of Hundreds of Millions of Yuan, Accelerating Buildout of Distributed Computing Power Network

    Gongji Technology announced the completion of a new round of funding totaling hundreds of millions of yuan. Investors include China Internet Investment Fund, Oriza Hua, and Saimu Technology, while existing shareholders Chunhua Venture Capital, Lihe Zhongke, and Xinjin Venture Capital continued to increase their investment. The funds will be used for technology iteration of the distributed computing power network, expansion of the industry ecosystem, and commercialization. The company has participated in the construction of computing power scheduling platforms in Qinghai, Hebei, the Guangdong-Hong Kong-Macao Greater Bay Area, and other regions, and has cumulatively served more than 12,000 enterprises and teams, with ARR exceeding 200 million yuan.

    #Industry news#English text is being prepared.#English text is being prepared.

  2. China.com Technology
    AI score24

    Development approach for a state-owned enterprise bidding and procurement document knowledge base agent: making bidding and procurement materials searchable, traceable, and auditable

    Wuhan Qichuang Dongli Digital Technology has proposed a development approach for a state-owned enterprise bidding and procurement document knowledge base agent: first conduct structured governance of regulation versions, project ledgers, and historical cases, then layer on hybrid retrieval combining vector search, keyword search, and structured field filtering, and move permission filtering ahead of retrieval. Its self-developed "Lingshu" AI agent PaaS platform turns knowledge governance, retrieval strategies, permission control, and audit trails into configurable underlying capabilities, and the question-answering chain can be fully exported for auditing.

    #Tutorials / practice#RAG#Agents

  3. Latent SpaceResearch
    AI score76

    LLMs Can Directly Read and Write Neural Network Weights: Weight Space Programming Proposes a New Gradient-Free Training Path

    The AI news roundup compiled by Latent Space introduced the Weight-Space Programming research by Takehi et al. Frontier LLMs can directly inspect and edit neural network weight tensors based on natural language goals, without gradient descent or training loops. Demonstrations include identifying the behavior of small models from raw weights, writing a small game MLP from scratch, editing 3.5% of the weights to repair a failed control policy, and editing 0.016% of the weights of Qwen3.5-0.8B to raise MATH-500 from 7.9% to 48.1%. Community skepticism holds that some results are closer to LLM-assisted interpretability analysis, that the MATH-500 improvement mainly comes from eliminating reasoning loops, and that the method did not work on Gemma 3 270M, which lacks such loops.

    #Papers/Research#English text is being prepared.#Open-source ecosystem

  4. Caixin
    AI score69

    Marvell plans to acquire Lumentum in a $115 billion all-stock deal

    Marvell announced it will acquire optical communications vendor Lumentum in an all-stock transaction valued at approximately $115 billion, a 23.9% premium to the October 28 closing price, with the deal expected to close in the first half of 2028. After the transaction is completed, Lumentum shareholders will hold approximately 33% of the combined company and Marvell shareholders approximately 67%, with Marvell Chairman and CEO Matt Murphy continuing to lead the combined company. On the day the news was announced, Marvell's stock fell about 13% and Lumentum's fell about 19%.

    Same story · 9 reports

    #Industry news#English text is being prepared.

  5. Phoenix TechnologyOpinion
    AI score60

    Apple's fiscal 2026 revenue reaches $478.8 billion as AI inference costs push up hardware pricing

    Apple released results for the fourth quarter and full fiscal year 2026, with full-year revenue reaching $478.8 billion and quarterly revenue of $114.477 billion, up 12% year over year. CEO Ternus said on the call that memory prices are rising at an unprecedented pace, putting direct pressure on hardware pricing, and gross margin guidance for next quarter falls back to 45%–46%. Siri handles more than 2.5 billion requests per day, and cloud inference spending pushed R&D expenses up to $45.765 billion, up 32.5% year over year; Apple is exploring bringing advanced AI capabilities into iCloud+ subscriptions.

    Same story · 9 reports

    #Industry news#OpenAI#English text is being prepared.

  6. Phoenix TechnologyOpinion
    AI score52

    Is Physical AI Heating Up or Cooling Down: Financing Cools While Tech Giants Enter

    Science and Technology Innovation Board Daily sorts out the hot and cold divergence in the physical AI sector: in 2026, primary market financing for physical AI and world models has exceeded 100 deals, but in October alone it dropped to 5 deals, with investors acting more cautiously. Big tech, meanwhile, is accelerating its layout of foundations and ecosystems: Google DeepMind released Genie 4, Nvidia expanded the Cosmos platform, and AMD acquired World Labs for about $8.2 billion. Most companies in the A-share industry chain remain in the position of shovel sellers; Suocheng Technology's physical AI business revenue in the first half of the year was 6.59 million yuan, accounting for 9.06%.

    #English text is being prepared.#Embodied AI#Industry news

  7. Phoenix Technology
    AI score23

    Joseph Tsai: AI can only be inclusive and accessible if it is open-source and open

    Joseph Tsai, co-founder and chairman of the board of Alibaba Group, said at the opening ceremony of the 2026 World Internet Conference Wuzhen Summit that AI can only be inclusive and accessible if it is open-source and open. He said that Chinese tech companies have continued to increase their efforts in open-sourcing models and opening up their ecosystems, maintaining a lead in key indicators such as open-source model downloads and the number of derivative models, and have become the current global leader in open-source intelligence. Open source can lower R&D costs for small and medium-sized enterprises, and its value is especially significant for minority languages and developing countries.

    Same story · 4 reports

    #English text is being prepared.#Open-source ecosystem

  8. Computerworld
    AI score47

    Can AI really help employees save time? Two sets of ActivTrak data contradict each other

    The latest ActivTrak research shows that heavy AI users work 8.42 to 9.08 hours per day, up to 90 minutes more than the 7.58 hours of those who do not use AI, contrary to its earlier conclusion this year that "AI shortens the workday by about 9 minutes." The new report is based on five months of behavioral data from 467 enterprises and 9,140 knowledge workers; heavy users' focus periods are 22% shorter than those of light users (37.8 minutes versus 48.2 minutes), and their app-switching frequency is 15% higher. Analysts point out that the recorded results are limited by the monitoring methods, and work completed via APIs or background agents does not generate desktop sessions.

    #English text is being prepared.#English text is being prepared.

  9. Sina Tech
    AI score45

    Morgan Stanley CEO: Hyperscalers May Spend $800 Billion to $1 Trillion Annually Over the Next 7 to 12 Years

    Morgan Stanley CEO Ted Pick said at the HKMA Global Financial Leaders' Investment Summit in Hong Kong that the largest hyperscalers may spend $800 billion to $1 trillion annually over the next 7 to 12 years, and Morgan Stanley estimates that cumulative AI infrastructure spending over the next 10 years could reach $15 trillion. Goldman Sachs CEO David Solomon said capital markets remain willing to absorb AI financing demand, and the main factor limiting deal size is the cost of capital rather than whether funds are sufficient.

    #Industry news#English text is being prepared.

  10. Sina Tech
    AI score35

    Alibaba's Joseph Tsai discusses AI development at Wuzhen: only open source and openness can bring inclusive, widespread access

    Joseph Tsai, Chairman of the Board of Alibaba Group, said at the opening ceremony of the 2026 World Internet Conference Wuzhen Summit that AI can only become widespread through open source, and only through widespread adoption can a sustainable, inclusive intelligent era be created. Alibaba's Qwen has open-sourced more than 470 models, with cumulative global downloads exceeding 4 billion and more than 330,000 derivative models, making it the world's number one open-source model family. The ModelScope community has attracted 33 million users from 239 countries and regions worldwide, with access to 280,000 open-source models, 40,000 datasets and 12,000 MCPs.

    Same story · 4 reports

    #English text is being prepared.#Open-source ecosystem#Agents

  11. The Paper
    AI score48

    NDRC responds to The Paper: closely monitoring AI's impact on employment structure, promoting the inclusion of new professions such as model trainers in the national occupational classification system

    Wang Ruomeng, deputy director of the Department of Innovation and High-Tech Development of the National Development and Reform Commission, said in response to a question from The Paper that the commission will closely monitor the impact of artificial intelligence on the employment structure, promote the inclusion of new professions such as model trainers and data annotators in the national occupational classification system, and strengthen monitoring and early warning of job displacement risks and improve social insurance linkages. During the "15th Five-Year Plan" period, the NDRC will focus on four areas of work, including accelerating the advancement of foundational and comprehensive legislation on artificial intelligence, supporting the establishment of AI "regulatory sandboxes," clarifying the responsibilities of multiple entities for collaborative governance, and addressing social impacts such as employment, rights protection, and mental health.

    Same story · 6 reports

    #English text is being prepared.#Industry news

  12. The Paper
    AI score38

    National Development and Reform Commission: Exploring new governance mechanisms for data security issues involved in the use of data for artificial intelligence

    Luan Jie, Director of the Policy and Planning Department of the National Data Administration, stated at the "15th Five-Year Plan" series press conference that new governance mechanisms will be explored for data security issues involved in the use of data for artificial intelligence, so as to both enhance data security protection capabilities and leave room for compliant innovation for new technologies and new products. The National Data Property Rights Registration Service System has been launched and put into operation; the next step will be to promote the role of registration certificates as credible proof in data transactions, asset capitalization, and judicial adjudication. In addition, the formulation of top-level documents for a nationally integrated data market will be accelerated, and a new round of the three-year "Data Elements ×" action will be planned and implemented.

    #English text is being prepared.#English text is being prepared.

  15. Google BlogFirst-party
    AI score31

    Google establishes a research hub in Singapore, providing over $6 million in research funding to NUS and NTU

    Google has established a research hub in Singapore and is investing more than $6 million in research funding in the National University of Singapore (NUS) and Nanyang Technological University (NTU), covering areas such as Alzheimer's disease research, computational efficiency and regional language processing, and AI safety. Of this, NUS receives a $1.5 million AI for Science Google.org Impact Challenge Grant, NTU receives $340,000 for four AI safety research projects, and there is an additional $750,000 AI for Science Capacity Fund Grant. Google.org and Google DeepMind are also providing $3 million to the INSEAD Singapore Thrive Lab.

    A light beam is focused through a lens into thin wires that pass through panels featuring increasingly complex patterns of blue dots, dashes, and other geometric shapes. The Google G is on the bottom left corner.

    #Industry news#Google#English text is being prepared.

  16. Leiphone
    AI score22

    World Internet Conference AI Committee Focuses on Agent Security, 360 Participates in Governance as a Core Member

    The 2026 plenary meeting of the World Internet Conference Artificial Intelligence Committee was held in Wuzhen. The draft 2027 work plan reviewed at the meeting proposes focusing on "controllable and collaborative governance of agent security," extending toward agent behavior control, governance capacity building, and global governance mechanisms. With the approach of "using models to govern models and using agents to counter agents," 360 launched an agent security protection and management system covering asset discovery, behavior control, threat detection, incident response, and loss recovery. Its self-developed AI security attack and defense system "Tulongfeng" passed the China Academy of Information and Communications Technology's "Internet Agent Governance—Vulnerability Discovery Agent Capability Assessment," becoming the first product to pass this assessment.

    #Industry news#Safety / alignment#Agents

  17. LeiphoneAnalysis
    AI score63

    REA Reverse-Engineering Test: Gemini 4 Argon and GLM 5.3 Recreate the Palace Museum Famous Paintings Chronicle

    Leiphone paired REA with Reverse-skill in OpenCode, having Gemini 4 Argon and GLM 5.3 separately reverse-engineer and recreate the artwork display feature of the Palace Museum Famous Paintings Chronicle. Gemini 4 Argon retained the original site's three-column layout, four filters, and the detail-page banner and related recommendations, while GLM 5.3 changed the three columns to four, removed the filter area, and replaced the detail page with a white content box, making it closer to a simplified browsing prototype. The article also cites the AA leaderboard intelligence index, where Argon's high tier scores 53 points, tying with the max tier of GPT-6 Astra and Fable 5.1.

    #Evaluations / benchmarks#Agents#English text is being prepared.

  18. Leiphone.comAnalysis
    AI score67

    Hands-on test of Grok Bot, WorkBuddy, and Qoder CN: Writing a large model routing review on the same topic

    Leiphone.com used the same set of instructions to have Grok Bot, WorkBuddy, and Qoder CN each assemble multi-agent teams to write "A Research Review of Large Model Routing." Grok Bot delivered in 12 minutes 49 seconds and included 19 papers, making it the fastest and broadest in coverage. WorkBuddy took 21 minutes 43 seconds, and its main text was closest to a research draft that could be edited directly; Qoder CN took 14 minutes 42 seconds, and its verification agent found that the source of "up to 85% cost savings" was unclear, and distinguished the evaluation basis for the 3.66x cost savings. The three deliverables each left gaps in file citations, numerical bases, and method classification, and the article accordingly argues that the next competition among multi-agent products is whether verification conclusions can pass through each handoff and land in the final file.

    #Evaluations / benchmarks#Agents#Multimodal

  19. Leiphone.comAnalysis
    AI score58

    Leiphone.com tests DeepSeek V4.1-Flash: Is the periodic weakness found by ByteDance Seed still present?

    Following the FP8 code completion experiment in the ByteDance Seed paper, Leiphone.com conducted two rounds of testing on the currently deployed V4.1-Flash via the official DeepSeek API. All 64 fixed samples were completed with the correct answer "8", and the answer flip between "8" and "32" reported in the paper did not occur. However, the model's predicted probability for "8" varied with padding length: in the docstring group, when padding increased from 24 tokens to 25, P(8) rose from 83.33% to 92.29%. Data from the Seed paper shows that for the V4-Flash base model, the accuracy gap between the best and worst phases was 40.2 percentage points; after post-training it narrowed to 19.1 percentage points, and for V4.1-Flash it further dropped to 6.1 percentage points.

    #Papers/Research#Reasoning#DeepSeek

  20. Leiphone.comOpinion
    AI score65

    AI Product Knockout: Big Tech Consolidates Agent Product Lines, Character.AI Files for Bankruptcy Reorganization

    Leiphone.com reviewed 16 representative AI products publicly launched this year by Tencent, Alibaba, ByteDance, and Baidu. As of October 19, 1 had announced service shutdown, and 5 had undergone integration or organizational adjustments; ByteDance's TRAE Work went from upgrade to being merged into the Doubao system in just 76 days. Tencent's QClaw announced in September that it would cease operations at year-end and migrate to WorkBuddy, while Alibaba integrated Wukong, QoderWork, and MuleRun into Qianwen Office. Character.AI, which has about 12 million active users, initiated Chapter 11 proceedings under the U.S. Bankruptcy Code on October 12; it loses $8 million to $9 million per month, spends about $10 million on compute, and its subscription monthly revenue peaked at about $2.8 million.

    #English text is being prepared.#Agents#Industry news