NAll news

All news

Wednesday916 items

  1. 36Kr · AIOpinion
    AI score55

    Is physical AI heating up or cooling down: primary market cools while big tech accelerates deployment

    A STAR Market Daily article republished by 36Kr sorts out the hot-and-cold divergence in the physical AI track: in 2026, primary-market financing for physical AI and world models has exceeded 100 deals, but in October alone it fell to 5 deals, with investors saying they have slowed their pace of investment. On the big tech side, Google DeepMind released Genie 4, AMD acquired World Labs for about $8.2 billion, Nvidia expanded the Cosmos platform, and OpenAI released GPT-6 Astra. Most A-share companies remain in the position of selling shovels; Suochen 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

  2. 36Kr · AIOpinion
    AI score53

    Megvii's 15th Anniversary: Zhao Kang's Internal Letter Discusses Strategic Convergence and Turning Profitable in 2026

    In the 15th anniversary internal letter, Megvii CEO Zhao Kang proposed strategic convergence, focusing on physical space operations, using POA OS as the foundation to implement five scenarios—urban governance, transportation operations, business intelligence, workplace safety, and workplace management—and expects to achieve full profitability in 2026. The internal letter disclosed that the total number of model releases in the first half of 2026 increased by 180% compared with the same period in 2025, and the contract amount in the business intelligence field is expected to increase by more than 100% year-on-year. The article also points out that Megvii had cumulative losses of 14.634 billion yuan from 2018 to the first half of 2021, and co-founders Yin Qi, Tang Wenbin, and Yang Mu have all left Megvii to start new projects.

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

  3. 36Kr · AIOpinion
    AI score60

    AI Takes US Stock Market's Magnificent Seven Hostage: In Earnings Season, Who Is Making Money Now and Who Is Betting on the Future

    An analysis by "Dingjiao One" republished by 36Kr points out that, as of November 2, the six US tech giants that had released earnings reports saw revenue growth across almost the board, but their stock price trajectories diverged: Google, Meta, and Amazon, which had negative free cash flow, fell, while Microsoft, Tesla, and Apple, which had positive free cash flow, held steady. The article breaks down each company's business model and argues that the arms race phase of AI infrastructure is giving way to a phase of crunching the numbers, with the next validation point being Nvidia's November earnings report.

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

  4. 36Kr · AI
    AI score27

    Former Claude Design Lead Jenny Wen Joins Musk's Camp, to Lead Grok Bot Design

    Jenny Wen, who previously led design for Claude and Cowork, has joined Musk's camp following SpaceX's acquisition of Cursor and is involved in the design of Grok Bot, which launched its Vault on October 23. She previously led directional design for Cowork as an independent contributor at Anthropic and was involved in launching Claude Dispatch.

    #Industry news#Agents

  5. 36Kr · AI
    AI score60

    OpenAI API briefly shows dual virtual machine model codename rc-ada, withdrawn after a few minutes

    Two undisclosed model IDs, rc-ada-or-vm and rc-ada-sh-vm, briefly appeared in OpenAI's official API model list and were taken down a few minutes later; during that window, developers who initiated calls encountered 503 capacity-limited errors. The article states that the rc prefix refers to a release candidate, ada is an internal codename, and the two vm suffixes correspond respectively to a brain virtual machine responsible for task scheduling and a virtual machine that executes commands in an isolated sandbox, and it cites the gpt-6.2-astra and gpt-6.2-betelgeuse codenames from internal Codex logs. In the BenchCAD 2.0 evaluation chart in OpenAI's blog post "Accelerating Engineering R&D in the Physical World" published the same day, an opponent labeled as an internal model scored 75.8% in its starting tier, higher than GPT-6 Astra's 59.4% in its full-power tier, with a peak of 87.5%.

    #Industry news#OpenAI#Agents

  6. 36Kr · AI
    AI score31

    Creator Xu Fei earns nearly one million yuan in revenue share on Hongguo with AI comic dramas, at a per-title cost of about 2,000–3,000 yuan

    Xu Fei, who entered the industry with zero foundation, has earned nearly 800,000 yuan in dividends on Hongguo through AI comic dramas, with a per-title production cost of about 2,000–3,000 yuan. His nearly ten short dramas have accumulated over a billion views in total, with the highest single title exceeding 100 million. He compressed the computing cost of a 50-minute drama to 600 yuan and 800 yuan for gacha draws, relying on Fanqie Novel's open scripts and rapid mass production to capture traffic. Starting November 1, Hongguo's new rules take effect, with completion incentives leaning more toward 90–180 minutes.

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

  7. 36Kr · AIOpinion
    AI score61

    Chinese embodied intelligence going global: from selling robots to open-source models entering overseas toolchains

    Of the 182 funded embodied intelligence companies counted by EO Intelligence, 73 have already gone global and 21 are preparing to, together accounting for about 51.6%. The article divides going global into two paths: asset-heavy product and supply chain expansion (Unitree, AgiBot, UBTECH, etc.) and technology platform ecosystem expansion, and cites Ant Lingbo's LingBot series as an example, saying it has more than 33,000 GitHub stars and over 2.2 million downloads, with projects such as Stanford's OpenWAM and the National University of Singapore's Velox reusing its model architecture or foundation.

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

  8. indigoOpinion
    AI score52

    indigo analyzes space data center cooling solution: aluminum radiators about 1300 W/m², bottleneck is Starship launch capacity

    Drawing on Elon Musk's reply and the public AI1 satellite design, indigo outlined a space data center solution centered on radiative cooling. An aluminum radiator facing deep space achieves about 1300 W/m², while solar panels produce about 250 W/m², so the solar panel area is about 5 times that of the radiator, and most of the satellite's mass comes from the solar panels. The AI1 satellite has peak compute of about 150 kW and average compute of about 120 kW, with deployable liquid radiators of about 110 m², a wingspan of about 70 m, and a deployed height of about 20 m; the author believes the real bottleneck is whether Starship can deliver large-area structures to orbit at low cost.

    Quoted postElon Musk · @elonmusk

    Yeah, much easier to cool in space. ~1300 W/m^2 is easily achievable for an aluminum radiator pointed away from the Sun, as it is mostly just seeing cosmic background radiation at ~3 degrees above absolute zero! Solar panels generate ~250 W/m^2, so you should expect them to have ~5 times the area of the radiator. Most of the mass of the satellite is solar panels.

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

  9. Electronic Technology Application
    AI score47

    NVIDIA: AI Reshapes Chip Factories, Simulation Speeds Up by Up to 35x

    NVIDIA says AI is driving semiconductor fabs into the "AI-defined factory" stage, accelerating chip manufacturing decisions through accelerated simulation, learned physics models, AI agents, and digital twins. Applied Materials' ACE+ platform uses PhysicsNeMo to achieve up to 35x acceleration in topography simulation, while Siemens Calibre, after integrating cuLitho, speeds up key simulation workloads by up to 17x. Applied Materials has also built a digital twin of an entire wafer fab based on NVIDIA Omniverse, simulating material flow and validating layouts before production begins.

    Same story · 2 reports

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

  10. China.com Technology
    AI score17

    How LibTV (Libu TV) Lets Beginners with No Editing Skills Complete AI Short Dramas

    LibTV (Libu TV) puts the script, characters, storyboards, images, video, audio, and editing onto a single infinite canvas, enabling beginners with no editing skills to go from script to finished film. Script nodes can automatically break down structured shots and extract characters, scenes, and props; storyboard images and video clips support shot-by-shot modification, and multi-track compositing supports exports of up to 20 minutes. The platform also provides an Agent entry point, calling models and orchestrating workflows through the Skill interface to automatically complete everything from script generation and storyboard design to video output.

    #Tutorials / practice#Video#Agents

  11. China.com Technology
    AI score25

    Want to make AI short dramas? Which AI video platforms are worth trying first: LibTV (Libu TV) one-stop platform and open-source tool comparison

    LibTV (Libu TV) is a one-stop AI video platform for AI films, advertising videos, and short drama creation. It uses an infinite canvas and node-based workflow to complete scripts, characters, scenes, storyboards, generation, and editing within the same project. The platform highlights Agent and Skill capabilities, allowing reuse of tuned workflows. The 3D director's console supports multi-camera, 9-grid and 25-grid, focus, and HD video shot controls, and will soon integrate the video model Seedance2.5. Open-source projects suit creators willing to configure environments themselves and customize as needed; beginners can first use a commercial platform to get the process working.

    #Tutorials / practice#Video#Agents

  12. China.com Technology
    AI score21

    How to Choose an AI Short Drama Tool: LibTV (Libu TV) Connects Storyboarding, Generation, and Editing with an Infinite Canvas and Node Workflow

    LibTV (Libu TV) organizes AI short drama production with an infinite canvas and node-based workflow, enabling scripts, characters, storyboards, image and video generation, and editing to be completed within the same project. It offers two views, Storyboard and Node Workflow; the 3D director's console supports multi-camera setups, 9/25-grid layouts, and batch generation of alternative shots, and allows shot-by-shot refinement and output of multiple versions by language and aspect ratio. The platform will soon integrate the video model Seedance2.5.

    #Product updates#Video#Agents

  13. China.com Technology
    AI score23

    How LibTV (Libu TV) uses a one-stop workflow to help beginners make AI short films

    LibTV (Libu TV) brings the script, characters, storyboard, generation and editing together in a single project through an infinite canvas and node-based workflow. Beginners can first arrange 6 to 8 shots in the Storyboard view, then refine them shot by shot. The platform offers controls such as a 3D director's stage, multi-camera and 9/25-grid layouts, and will soon integrate the video model Seedance2.5; the Agent takes on repetitive work such as batch generation and asset replacement through Skills.

    #Tutorials / practice#Video#Agents

  14. China.com Technology
    AI score25

    TCL Industries Selected for World Economic Forum MINDS AI List, Pan-Smart Screen Intelligent Container Computing Cabinet Wins AI Application Star

    The TCL Industries Pan-Smart Screen Intelligent Container Computing Cabinet team was selected for the MINDS list announced by the World Economic Forum in June 2026, winning the "MINDS AI Application Star," and its case was included as a benchmark example under "Building Resilient Infrastructure and Supply Chains." The project uses AI to optimize container loading planning, shortening single-container planning time from 30 minutes to seconds—a 5,900% efficiency improvement—saving over 10 million in costs annually and reducing carbon emissions by 762 tons. In the past three years, TCL's AI applications have created comprehensive benefits exceeding 2.5 billion yuan, and the Star Intelligence large model ranked 11th in the 2025 global industrial large model rankings and first in the display field.

    #Industry news#English text is being prepared.

  15. China.com Technology
    AI score22

    AI Short Drama Production Tool List: LibTV One-Stop Platform and Open Source Project Comparison

    LibTV (Libu TV) covers the complete workflow from script, characters, and storyboards to image and video generation and editing with an infinite canvas and node-based workflow, and highlights its Agent and Skill capabilities, with the video model Seedance2.5 soon to be integrated. Its 3D director's console supports multi-camera, 9-grid and 25-grid, focus, HD video and other shot controls, enabling batch generation of multiple shots followed by shot-by-shot fine-tuning. The article also organizes locally deployable open source projects by production stage, for creators willing to debug on their own to supplement their choices.

    #Tutorials / practice#Video#Agents

  16. China.com Technology
    AI score37

    BooleanMix's Four Rounds of Iteration: Using the TOLD Tech Stack to Solve the Challenge from Heterogeneous Computing Power to Token Delivery

    Through four rounds of iteration on the BooleanMix platform, BooleanMix has built capabilities around the unified management of heterogeneous computing power and Token delivery. Its self-developed TOLD tech stack has completed real-world validation on a 10,000-GPU-scale heterogeneous cluster of NVIDIA and domestic GPUs. A report by the China Academy of Information and Communications Technology points out that inference computing power has surpassed training computing power for the first time to become the main source of demand, and that Tokens are replacing FLOPS and GPU card counts as the new yardstick for measuring the value of the AI industry. BooleanMix is simultaneously opening up two types of partnership directions, working with IDC operators, intelligent computing centers, and investors to advance large-scale delivery from computing power assets to Token production capacity.

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

  17. Smart City Industry Analysis
    AI score26

    Li Qiang's keynote speech at the 2026 World Internet Conference Wuzhen Summit: Building a community with a shared future in cyberspace

    In his keynote speech at the opening ceremony of the 2026 World Internet Conference Wuzhen Summit, Li Qiang stated that internet development is undergoing an intelligent transformation and systemic leap, and that the internet should be reunderstood from three dimensions: infrastructure, the driving force of industrial transformation, and the power of intelligent generation. He emphasized that internet development must highlight a people-centered approach, responding to four issues: humans and machines, the ethics of algorithmic models, the security of the integration of data and the real economy, and fairness regarding the digital-intelligent divide, and he proposed working together to build a community with a shared future in cyberspace from four aspects: technology empowerment, trustworthiness and controllability, inclusiveness and balance, and open dialogue.

    #English text is being prepared.#Industry news

  18. East MoneyProduct release
    AI score58

    CrowdStrike Launches AI Kill Switch on Falcon Guardian Platform

    CrowdStrike announced the launch of an AI kill switch in the Falcon Guardian platform, which allows security personnel to terminate the operation of enterprise AI agents in real time, prevent them from restarting, and cut off their access to the corporate network when they deviate from expectations, behave maliciously, or go out of control. The feature is similar to a circuit breaker; after agent activity is detected, a human still decides whether to cut off and restore access. CrowdStrike also announced plans to acquire AI observability startup Fiddler AI and AI security testing company Dreadnode.

    Same story · 2 reports

    #Product updates#Safety / alignment#Agents

  19. Phoenix Technology
    AI score41

    Harvard's youngest Chinese full professor Xi Yin leaves OpenAI after only 5 months

    Xi Yin, a string theorist and the youngest Chinese full professor in Harvard's history, has announced his departure from OpenAI. From reports of his joining in late May this year to his departure, roughly half a year passed. He said this was his first corporate job and that he will explore new directions. Yin entered the University of Science and Technology of China's youth class at age 12, was promoted to full professor at Harvard at age 31 in 2015, and has received the Sloan Research Fellowship and the 2017 New Horizons in Physics Prize; he previously estimated that, with extensive human supervision, AI sped up his research coding work by at least 100 times.

    Same story · 3 reports

    #Industry news#OpenAI

  20. Phoenix Technology
    AI score67

    Report: ByteDance Restricts Employee Access to Internal Materials via External AI Tools Such as Claude and Codex

    According to a report by Dachang Ribao, ByteDance has introduced new internal rules on the use of AI tools. Starting this month, external AI access is banned for all L4-level code repositories and Feishu documents company-wide, and AI accounts purchased by employees themselves will no longer be able to access the relevant internal resources. ByteDance executive Hong Dingkun responded on the internal network that internal information security is an important consideration behind this adjustment, and that the consequences of any leak of L3-and-above content would be serious.

    Same story · 2 reports

    #Industry news#Safety / alignment#English text is being prepared.