NAll news

All news

Monday595 items

  1. The Decoder
    AI score44

    OpenAI Prepares Lobbying Offensive in Washington After Midterm Elections

    OpenAI is preparing political lobbying and public relations efforts for after the U.S. midterm elections, in order to address potentially heightened regulatory pressure after Democrats win the House. Policy chief Chris Lehane plans to make "AI does not replace people" the core message, promote binding safety rules, and call them "speed limits." OpenAI also plans to rename data centers as "AI Compute Campuses" and promise economic benefits for the communities where they are located; its top priority in the new Congress is to accelerate the advancement of safety rules for frontier models.

    Same story · 10 reports

    #English text is being prepared.#OpenAI

  2. Phoenix TechnologyModel releases
    AI score78

    Meituan LongCat Team Releases 1.6-Trillion-Parameter Natively Multimodal Model LongCat-2.5

    The Meituan LongCat team has released LongCat-2.5, a natively multimodal model designed for long-horizon agent tasks, with a total parameter count of 1.6 trillion and approximately 48 billion parameters activated per Token. On benchmarks published by Meituan such as AutomationBench and LongCatClawBench, LongCat-2.5 leads Claude-Opus-5 and GPT-5.6-Sol; on Terminal Bench v4.0 it scores 38.5, surpassing GPT-5.6-Sol but trailing Claude-Opus-5. The model converts visual signals into multi-level discrete Tokens and processes them uniformly with text, and it redesigns its parallelism strategy for domestic accelerators whose single-card memory is under 80GB, raising single-card throughput for long-sequence SFT by 86% over the DDP baseline.

    Same story · 11 reports

    #Model releases#Agents#Multimodal

  3. Phoenix TechnologyProduct release
    AI score62

    Karpathy Built the HackerTube Video Recommendation Website with Claude Code in Two Hours

    Andrej Karpathy, former Tesla AI director and OpenAI co-founding member, released a weekend project called HackerTube. Dissatisfied with YouTube's recommendation mechanism, he used Anthropic's Claude Code to filter 6,454 repeatedly recommended videos from 43 million Hacker News comments spanning the past 20 years, and built a new video recommendation website. The whole process took about 2 hours. The project used the Claude Opus 5.5 model with reasoning effort set to Medium. Karpathy said he personally completed only three operations: creating a Google Cloud API key, starting a VPS server, and purchasing a domain name; the rest of the development, deployment, and security hardening was done by Claude. The website requires no registration or login, uses thumbnails taken from the middle of videos rather than creator covers, and provides a ZIP archive of the underlying comment analysis data for other developers to adapt on their own.

    Same story · 8 reports

    #Product updates#English text is being prepared.#Agents

  4. Sina TechResearch
    AI score60

    Tsinghua, together with Infinigence AI and AgiBot, proposes Collective Experience Learning (CEL); validation on 96 humanoid robots achieves a 97.5% success rate

    Tsinghua University, together with Infinigence AI and AgiBot, proposes a new paradigm called Collective Experience Learning (CEL), enabling robots to share successful experiences and failed lessons among themselves, and completes validation on 96 real humanoid robots. Across five tasks—flower arranging, clothes folding, phone charging, stool assembly, and phone packaging—the collective overall success rate reaches 97.5%, all exceeding the average level of 6 human operators, with the success rate on the phone packaging task rising from 79.2% to 98.3%. The experiment also shows that after the number of robots participating in training expands from 16 to 96, overall capability further improves; at the system level, a fault-tolerant asynchronous training system, FATS, is adopted, with time loss caused by hardware failures accounting for only 1.3% of total runtime, and at the algorithm level, the FERT experience retrieval fine-tuning algorithm is adopted.

    Same story · 9 reports

    #Papers/Research#Embodied AI#English text is being prepared.

  5. Sina Tech
    AI score43

    Unitree Robot G1 Withdraws from the First CMG World Robot Mount Tai Climbing Competition Due to Overheating, Carried Down the Mountain on a Stretcher

    During the first CMG World Robot Mount Tai Climbing Competition, the Unitree Robot G1 development version experienced an issue due to machine overheating. For safety reasons, the operator chose to withdraw from the race, and staff placed it on a stretcher and carried it down the mountain. The course is 3.9 kilometers long with more than 3,100 steps. The event includes four official competition categories—humanoid robot racing, legged robot racing, legged robot load-bearing, and exoskeleton robot assistance—as well as one all-terrain mobile pathfinding challenge.

    Same story · 3 reports

    #Industry news#Embodied AI

  6. Sina Tech
    AI score22

    LianShang Group Exhibits at the 2026 "Internet Light" Expo, Showcasing Multiple AI Products Including Xuqu and Haoling AI

    At the 2026 "Internet Light" Expo, LianShang Group showcased multiple AI products, including Xuqu, a zero-barrier music creation community; Haoling AI, aimed at the senior population; Zaolang, an AI creator learning platform; and Kouxin, a social product built on multimodal large models and AI agents. The Kouxin AI intelligent social platform and the Tandao AI-native game innovative product matrix were shortlisted for "New Light Products," while the Haoling AI smart senior living application solution was shortlisted for "New Shine Scenarios."

    #Industry news#Multimodal#Speech

  7. Sina Technology
    AI score40

    The first CMG World Robot Mount Tai Climbing Competition concludes, with Unitree Technology winning 1 gold and 2 silver

    The first CMG World Robot Mount Tai Climbing Competition concludes, with Unitree Technology taking 1 gold and 2 silver: the Beijing Lingyi Technology team used a Unitree A2 robot dog to win the legged robot speed race championship with an actual time of 41 minutes 41 seconds (43 minutes 41 seconds after a 2-minute penalty), in fully autonomous mode throughout. The Shanghai Gaoyi Technology team added two more silvers: the Unitree A2 took second place in the speed race, and the Unitree B2, carrying an effective payload of 66 kg (74 kg including the rack) on the Wanxian Tower–Doumu Palace section, took second place in the payload race. The course runs from Hongmen to Zhongtianmen, covering 3.9 kilometers and more than 3,100 steps.

    Same story · 3 reports

    #Industry news#Embodied AI

  8. Shanghai Observer
    AI score20

    The Second Shanghai Senior Learning Course Exhibition opened, releasing courses including "AI Literacy 100 Points" (Senior Edition)

    The Second Shanghai Senior Learning Course Exhibition opened at the Shanghai Exhibition Center, with 102 enterprises and institutions exhibiting and displaying more than 500 courses, and releasing the first batch of 20 Shanghai senior education premium courses and "AI Literacy 100 Points" (Senior Edition). The latter features seven major sections with a total of 100 episodes, each about 5 minutes long, and adopts large fonts, high contrast, and AI video generation technology, and can be viewed on platforms such as "One Net Learning". Data shows that 69.6% of surveyed seniors are willing to participate in AI learning.

    Same story · 3 reports

    #Industry news#English text is being prepared.

  9. Shanghai Observer
    AI score50

    NRTA Cracks Down on "AI Tampering": Over 11,000 Violating Videos Cleared and More Than 30 Accounts Handled in October

    The National Radio and Television Administration disclosed that a special governance campaign against "AI tampering" videos on some key online audiovisual platforms in October cleared more than 11,000 violating videos and handled more than 30 violating accounts. The governance effort focused on "AI tampering" videos based on classic film and television works such as the Four Great Classical Novels, historical themes, revolutionary themes, and heroic model figures, while also clearing various types of cult animations. The special governance campaign was launched in January 2026 and shifted to a normalized, institutionalized long-term mechanism starting February 1.

    Same story · 2 reports

    #English text is being prepared.#Video

  10. Shanghai Observer
    AI score32

    The Unitree G1 robot featured in the Spring Festival Gala overheated and withdrew midway while climbing Mount Tai, and was carried down the mountain on a stretcher

    In the first CMG World Robot Mount Tai Climbing Competition, the Unitree G1 developer edition featured in the Spring Festival Gala withdrew due to overheating, with the operator choosing to pull out for safety reasons, after which staff carried it down the mountain on a stretcher. The course is 3.9 kilometers long with more than 3,100 stone steps, and other robots also stumbled or needed to be pulled by ropes; robot number E04 became the first contestant to reach the finish line at Hutiange.

    Same story · 3 reports

    #Industry news#Embodied AI

  11. AYi AI Notes
    AI score30

    Jensen Huang responds to CNN: AI safety is like the brakes on a race car

    In a CNN interview, Zakaria pressed Jensen Huang on whether the AI revolution will ultimately leave humanity safe and sound. He responded from an engineering perspective: electricity brought fires, aviation brought plane crashes, and taming powerful machines depends on constant vigilance rather than guarantees of absolute harmlessness. He rarely publicly thanked Turing Award winner Hinton for his warnings, and believes that safety and the advancement of computing power should receive equal effort—safety is the brake that lets the driver dare to press the accelerator hard, not a stumbling block.

    Quoted postAYi · @AYi_AInotes

    English text is being prepared.

    #English text is being prepared.#Safety / alignment

  12. Rohan PaulResearch
    AI score85

    Epoch AI test: Zhipu GLM 5.3 turned unreleased Firefox security fixes into working attacks across 6 consecutive versions

    In Epoch AI's tests, Zhipu's open-weight model GLM 5.3, across 6 consecutive Firefox versions, identified publicly available security fixes that had not yet reached the stable release as real patches and independently built complete attack chains, with no refusals throughout. The resulting web pages could send visitors' local files to the attacker without any additional clicks, and could reach saved passwords and cookies. The research team first invested more than $100,000 to build the security testing system; after that, each attack cost about $1,000, with the fastest taking 12 hours; each attack also used a second unknown Firefox vulnerability that the AI discovered on its own, which Mozilla fixed on October 27. A similar attempt against Google Chrome did not succeed even after spending about $100,000.

    #Safety / alignment#Agents#Open-source ecosystem

  13. Leiphone
    AI score48

    Inside the first batch of FDEs' survival stories: some landed ten-million-yuan deals, others were rejected at a 3,000-yuan quote

    The first batch of FDEs (Forward Deployed Engineers) has fallen into pricing chaos in the AI implementation market: a cross-border e-commerce restocking forecast project shrank from the million-yuan level to 100,000 yuan, enough only to buy 10 WorkBuddy accounts plus data interface modifications; another supply chain project for a multinational tech giant was quoted at 8 million yuan, more than ten times higher than similar projects. Entrepreneur Gao Feng was strung along more than a dozen times in two months, and one FDE turned down an order because quoting only 3,000 yuan per requirement could not cover token costs. Some software companies still use SaaS middle-platform and person-day pricing, making FDEs prone to becoming on-site outsourced staff or "freebies."

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

  14. LeiphoneProduct release
    AI score60

    Alibaba Cloud Releases Qovest, an AI Investment Research Workbench

    Alibaba Cloud has released Qovest, an AI investment research workbench for professional institutional investors. Users submit a research request in a single sentence, and it automatically queries data, collects materials, analyzes them, and generates a report; it can also continuously track research subjects at a set frequency. Qovest consolidates the results of each research session into a research space, allowing research to be carried forward; the team procures authoritative financial databases and has developed its own Agent engineering framework for financial scenarios, Harness. The official website is currently open for invitation-only testing for individual users, with subscription plans of 99 yuan or 199 yuan per month to be offered later, and an enterprise edition to be launched soon.

    Same story · 4 reports

    #Product updates#Agents#RAG

  15. LeiphoneModel releases
    AI score70

    Meituan Releases LongCat-2.5 Large Model, 1.6 Trillion Parameters Trained Entirely on Domestic Compute

    On November 2, Meituan released its new-generation large model LongCat-2.5, a native multimodal model designed for long-horizon agent tasks with 1.6 trillion total parameters. The model continues the research direction of the LongCat series and was trained entirely on domestic compute, making it the largest training task completed on domestic compute to date.

    Same story · 11 reports

    #Model releases#Multimodal#Agents

  16. Leiphone
    AI score53

    Details of DeepSeek's second funding round of 105.5 billion yuan revealed, with vivo, Meituan, Xiaohongshu and others among investors

    DeepSeek's second funding round has raised 105.5 billion yuan (approximately US$15.8 billion), attracting nearly 30 investors, with state-owned capital contributing about 27%, a sharp increase from 2% in the first round. Tencent Holdings and CATL remain the two major investors; Tencent contributed 10 billion yuan in each of the two rounds, while CATL and its chairman Zeng Yuqun contributed 5 billion yuan and 10 billion yuan in the two rounds respectively. Other industrial investors also include ChangXin Memory Technologies, SMIC, vivo, Meituan, Geely Automobile, Midea Group, Luxshare Precision, Ctrip and Xiaohongshu, among others, with some companies investing between 1 billion and 3 billion yuan each.

    Same story · 5 reports

    #Industry news#Agents#DeepSeek

  17. Leiphone
    AI score49

    Hua Ying Aviation HE-1A First Production Aircraft Rolls Off the Line in Weinan, Ton-Class Commercial Unmanned Transport Aircraft Enters Mass Production

    The first production aircraft of the Hua Ying Aviation HE-1A commercial unmanned aerial transport system rolled off the assembly line at the Weinan final assembly base, marking the transition of China's ton-class commercial unmanned transport aircraft from a development model to mass production. The HE-1A has a maximum takeoff weight of 4 tons and a payload capacity of 2 tons, with cumulative signed orders of 110 aircraft; once all three production lines at the Weinan base reach full capacity, annual output will exceed 100 aircraft. The aircraft is scheduled to complete its maiden flight by the end of the year, and the company aims to obtain the type certificate before the end of 2027.

    #Industry news#Embodied AI

  18. Yicai
    AI score38

    Du Xiaoman CEO Zhu Guang: AI Moves Financial Services from "Tailored to Thousands" to "Tailored to Each Individual," Jointly Releases XuanYuan 4.5 Financial-Enhanced Large Model

    Du Xiaoman CEO Zhu Guang said at the Hong Kong FinTech Week that AI is moving financial services from "tailored to thousands" to "tailored to each individual," and that the biggest change in large models this year is the shift from "answering questions" to "delivering results." Du Xiaoman, together with the Hong Kong University of Science and Technology and the Hong Kong Generative AI Research and Development Center (HKGAI), jointly released the XuanYuan 4.5 financial-enhanced large model, which delivers results 30% better than general-purpose models in core financial scenarios such as customer acquisition, risk control, and customer management, at only 10% of the cost. Its risk control system's risk anomaly attribution plus strategy development cycle has been shortened from about two weeks to three to four days.

    Same story · 2 reports

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

  19. Yicai
    AI score50

    NRTA Cracks Down on "AI Tampering": Over 11,000 Violating Videos Cleared and More Than 30 Accounts Handled in October

    Some key online audiovisual platforms have released the results of the October special campaign against "AI tampering" videos, clearing a total of more than 11,000 violating videos and handling more than 30 violating accounts. Since February 1, the National Radio and Television Administration has shifted the governance effort into a normalized, institutionalized long-term mechanism, focusing on clearing violating videos based on classic film and television works such as the Four Great Classical Novels, historical themes, revolutionary themes, and heroic model figures, while simultaneously clearing various types of cult animations.

    Same story · 2 reports

    #English text is being prepared.#Video

  20. Tencent HunyuanFirst-partyProduct release
    AI score52

    Tencent Hunyuan Releases Research Agent Hyra 2.0

    Tencent Hunyuan has released Hyra 2.0, the next-generation version of the Hunyuan Research Agent, positioned to shift from solving problems to building reusable tools, models, and workflows in order to accelerate subsequent discoveries. The official team also published its views on the next stage of AutoResearch, with the link https://t.co/u2JSslKYai.

    Same story · 3 reports

    #Product updates#Agents#Reasoning