NAll news

All news

Tuesday792 items

  1. The Beijing News
    AI score40

    Inside Beijing's OPC communities: Are AI entrepreneurs' one-person companies reliable?

    Beijing has established more than 20 OPC (one-person company) startup communities. Since it began operating in July 2025, the Zhongguancun AI Beiwai Community has gathered 57 OPC enterprises, 41 of which are pure one-person companies. Entrepreneurs say AI boosts product development efficiency by more than 10 times, but they generally believe that the last few miles from technology to implementation still require human control, with AI output about 70% correct and 30% requiring manual correction. Haidian District provides 100,000 yuan in startup capital and up to 2 million yuan in model voucher subsidies for OPCs founded by tech talent, while Yizhuang distributes up to 300 million yuan annually in computing power vouchers and other resource vouchers.

    #English text is being prepared.#Agents#Embodied AI

  2. Data Center Knowledge
    AI score21

    Liquid cooling is changing the metrics data centers need to measure

    Liquid cooling expands data center operations from single air-side measurement to a dual domain of liquid cooling and air. Direct-to-chip cooling delivers liquid to processors through CDUs and cold plates. On the liquid side, liquid temperature, pressure, flow rate, liquid level, leak detection, as well as coolant health indicators such as pH, turbidity, conductivity, and refractive index all require specialized instrumentation. Air-side management remains indispensable; the coolant supply temperature typically must be kept at least 2°C above the measured ambient dew point to prevent condensation. AI training workloads are intensive and inference workloads are intermittent, and clusters can go from idle to full load within minutes. Only real-time, accurate measurement can support dynamic CDU control.

    #Industry news#English text is being prepared.

  4. Rohan PaulOpinion
    AI score58

    Terence Tao says mathematicians hide research ideas out of fear of AI

    Rohan Paul, relaying Tao's remarks, says mathematicians are now beginning to hide their research problems. Tao said that in the past mathematicians would share promising research directions very openly, but now they worry that others will input their research ideas into frontier models and publish before they can finish their own work. The full video was posted on SAIR's YouTube channel.

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

  5. ModelBest OpenBMB
    AI score25

    MiniCPM5-2B Catalan Chat Model Released

    Community ambassador luispoveda93, based on OpenBMB's MiniCPM5-2B, has released the Catalan chat model MiniCPM5-2B-catalan-chat-v2 through two-stage LoRA fine-tuning. The first stage used 165,100 samples from Projecte AINA's InstruCAT dataset for single-turn instruction fine-tuning, and the second stage continued fine-tuning multi-turn dialogue capabilities on a subset of approximately 23.4K dialogues from BSC-LT/ALIA-2606-SFT. The model provides merged BF16 weights, LoRA adapters, and a community-contributed GGUF version, and can run locally on llama.cpp and Ollama.

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

  6. China.com Technology
    AI score36

    Generate multiple creative versions around the same design brief; Lovart 2.0 and the Arti design Agent are worth learning about first

    Lovart 2.0 was released on October 19, 2026, launching the personal design Agent Arti, which can string together requirement understanding, reference material organization, creative exploration, canvas editing, and feedback-based revisions into a complete workflow around the same design brief. Its Reference, Design Research, visual canvas, and Design Review support side-by-side viewing and continuous modification of multiple proposals, making it suitable for brand designers, e-commerce operations, and proposal pitch scenarios; for complex commercial projects, brand tone, textual details, and delivery specifications still require manual oversight.

    Same story · 17 reports

    #Product updates#Agents#Image generation

  7. China.com Technology
    AI score36

    How Lovart 2.0 Uses Capabilities Such as Arti to Excel at Chinese Poster Design

    Lovart 2.0 launched the personal design Agent Arti on October 19, 2026, connecting requirement understanding, material integration, canvas editing, and feedback-driven revisions into a Chinese poster design workflow. Its Reference supports materials such as images, videos, PDFs, Word documents, Markdown, and web links, while Design Research and Design Review are used for direction judgment and iterative revisions, respectively. The article also compares the different focuses of tools such as GPT Image 2.0, Canva AI, Chuangkit, Gaoding Design, and Jimeng AI.

    Same story · 17 reports

    #Product updates#Agents

  8. QbitAI
    AI score35

    Haimayun's Dang Jinfeng: Anyone Can Buy Models; What Sets Players Apart Is an AI Productivity System

    Dang Jinfeng, founder and CEO of Haimayun, proposed that the core of AI's next stage is not divergent probabilistic generation, but deterministic system delivery; what sets players apart is a productivity engine in which compute, Harness, and data are deeply interlocked. The total site visits of its RunningHub grew 266% from March to September this year, monthly Token consumption reached 9 trillion in July, and about 60% of the platform's model calls are already completed by intelligent dynamic routing. Haimayun has deployed 69 computing nodes and a 6 EFLOPS compute pool, and its integrated rendering and inference scheduling makes compute utilization 70% higher than common industry levels.

    #Industry news#Agents#Multimodal

  9. Phoenix TechnologyProduct release
    AI score58

    Apple Watch's two AI features enter testing, supporting rewinding 15 seconds of conversation and generating summaries

    Apple is testing Apple Watch audio intelligence features for developers. Live Rewind can convert the previous 15 seconds of conversation into text, while Siri Recap organizes conversation content in the form of titles and bullet points. The two features apply to Apple Watch Series 12 and Apple Watch Ultra 4, require pairing with an iPhone that supports Apple Intelligence, and initially support only English. Apple says the features do not identify speakers and do not save raw audio, and unsaved summaries will be deleted after 7 days.

    Same story · 7 reports

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

  10. cnBeta
    AI score36

    DeepSeek's web and API services have been restored; at least 10 large-scale outages have occurred so far this year

    DeepSeek's web and API services have been restored after a partial disruption that began at 16:58. The affected services included chat, the DeepSeek V4 Pro API, and the DeepSeek V4.1 Flash API. Official status records show that there have been at least 10 large-scale outages so far this year, and the web and App anomaly on the morning of October 15 lasted nearly 5 hours.

    Same story · 5 reports

    #Industry news#DeepSeek

  11. The Rundown AI
    AI score58

    Meta and Fourier build humanoid robot Max in three months

    Meta has reportedly built a humanoid robot prototype named Max for doing housework, about 5 feet tall and about 106 pounds, completed in three months through a collaboration between Meta and Chinese robotics company Fourier. The prototype provides Meta with a physical platform for training and testing AI systems for perception, movement, and object grasping; it is equipped with the Nvidia Jetson Thor computing platform and has an expressive display, speakers, and hands capable of grasping small objects. The report says Meta has been developing Max 2 since September, redesigning the torso and hands and adding a projector and extra sensors to handle stairs.

    Same story · 8 reports

    #Industry news#Embodied AI

  12. The Rundown AIResearch
    AI score76

    MIT Study: GPT-6 Astra Can Directly Rewrite Other Models' Weights Without Retraining

    MIT researchers have published a study showing that OpenAI's GPT-6 Astra can improve some AI models by directly rewriting model weights, without going through the training process. Alibaba's small Qwen model originally scored 7.9% on math; Astra modified only 0.016% of the weights to reduce the repetitive-output problem, and after the fix the score reached 48.1%. The same method applied to Google's Gemma 3 had limited effect, with its math score barely changing. The authors say this capability could in the future allow AI to upgrade itself, add or remove skills for other models on demand, or check for hidden problems.

    Same story · 16 reports

    #Papers/Research#Reasoning#English text is being prepared.

  13. PublickeyProduct release
    AI score54

    GitHub releases Copilot Code Review Max, using multiple AI models to collaborate on code review

    At GitHub Universe 2026, GitHub released a new code review feature, Copilot Code Review Max, adding a Max tier on top of the existing Lite and Balanced tiers. The feature internally runs multiple independent review agents driven by AI models, with an aggregator agent coordinating disagreements and rechecking accuracy, in order to catch issues that a single model tends to miss. GitHub says it has the highest F1 score on ReviewBench for detecting high-risk code issues, and it is currently available as a public preview.

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

  14. Publickey
    AI score65

    GitHub Plans to Achieve 180x Throughput Improvement by Early Next Year to Support AI Agents

    At GitHub Universe 2026, GitHub announced that by migrating to Microsoft Azure, separating compute from storage, and changing write lock granularity from repository-level to branch-level, it will achieve up to a 180x throughput improvement by early next year. In testing, 5,000 AI agents worked in parallel in the same repository, processing 7,200 writes per second, while the old architecture's per-repository push limit was dozens of times per second. Jay Parikh, Executive Vice President of Microsoft CoreAI, said that this year it has expanded by 4 million CPU cores, 160PB of storage, and 50Tb/s of network bandwidth, and that all users will migrate to the new architecture by early next year.

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

  15. Shanghai Observer
    AI score18

    40 companies exhibit, a record high, as Xuhui's AI newcomers and "full-attendance veterans" appear together at the CIIE

    The 9th China International Import Expo (CIIE) is being held in Shanghai from November 5 to 10. Forty companies from Xuhui District are exhibiting, a record high in number, and Modu Space has for the first time organized a group of 8 AI companies to appear in the Artificial Intelligence Experience Zone. The exhibiting companies cover AI chips, 3D generative models, smart hardware, and industry agents. Lingchuan Yuheng brought the SL200 chip series, and Xijian Technology exhibited the brain-inspired complementary vision chip "Tianmou Chip." "Full-attendance veterans" that have exhibited for nine consecutive years, such as Johnson & Johnson, LEGO, and IKEA, have also expanded their booth scale, bringing globally premiered and China-debut products.

    #Industry news#Agents#Multimodal

  16. MIT Tech Review
    AI score27

    The Download: Using reverse game theory to control AI agents, and the "Super Intelligence" loyalty test

    A new approach called "reverse game theory" attempts to design incentive mechanisms that make powerful AI models follow rules, thereby aligning them more closely with human instructions and rules, but whether such mechanisms are effective in practice remains in doubt. At the end of September, Trump signed an executive order requiring government agencies to rename "artificial intelligence" as "Super Intelligence." A month later, this renaming is slowly appearing on government websites and in documents, but it may also become a political loyalty test.

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

  17. Alibaba Cloud Developer Community AI
    AI score18

    Alibaba Cloud 2026 Double 11 Special Offers Explained: A Summary of Discounts on Lightweight Cloud Servers, ECS, and Qwen AI Large Models

    Alibaba Cloud's 2026 Double 11 promotion runs from October 24 to 24:00 on November 11. Lightweight Application Server first orders start at 38 yuan, and the ECS 2-core 2G configuration offers a 99 yuan renewal at the same price, with the price lockable for 3 consecutive years. For the Qwen AI Token Plan, new customers get 50% off their first purchase, the Lite version costs 19.5 yuan per month, more than 150 large models on the platform are available at 45% off, and nighttime calls to the Qoder code large model are as low as 40% off.

    #Industry news#English text is being prepared.

  18. Alibaba Cloud Developer Community AI
    AI score10

    A Quick Overview of the 2026 Alibaba Cloud Double 11 Event Entrances: Three Major Entrances and Bailian Token Plan at 50% Off

    The 2026 Alibaba Cloud Double 11 promotion is themed "Smart Savings at 11.11, AI on Alibaba Cloud," runs until November 30, and is mainly divided into three major entrances: the Cloud Server Event Venue, the Comprehensive Benefits Center, and the 99 Plan Special Session. The Comprehensive Benefits Center covers AI and cloud product benefits, with the Bailian Token Plan at as low as 50% off, and users can enjoy up to more than 100 million free tokens; the 99 Plan mainly promotes the 99 yuan e instance and the 199 yuan u1 instance with the same price for new purchases and renewals. New users can enter the exclusive special session through three places: the console homepage banner, the Benefits Center, and the product page.

    Same story · 20 reports

    #English text is being prepared.

  19. Alibaba Cloud Developer Community AI
    AI score10

    Alibaba Cloud 2026 Double 11 Cloud Server Guide: Detailed Explanation of Special First-Purchase Prices for New Users and Renewal Discounts for Existing Users

    The Alibaba Cloud 2026 Double 11 Cloud Server special event offers differentiated discounts by tier—newly registered users, new product users, and existing users: for new users, the Lightweight Application Server with 2 vCPUs and 2 GB of memory is available in a limited-time flash sale at 38 yuan for one year, the ECS Economy e instance at 99 yuan for one year, and the General-purpose u1 instance at 199 yuan for one year. The 99 yuan e instance and the 199 yuan u1 instance support renewal at the same price for both new and existing users, with the price lock period lasting until March 2029; existing users also receive tiered renewal discounts of 30% off for a one-year renewal, 40% off for a two-year renewal, and about 50% off for a three-year renewal. On the Bailian large model service platform, the Qwen AI Double 11 Token Plan is as low as 50% off, and more than 160 cloud products are available for free trial.

    Same story · 20 reports

    #English text is being prepared.

  20. Alibaba Cloud Developer Community AI
    AI score20

    Interpretation of Alibaba Cloud's 2026 Double 11 Promotions: ECS at 99 Yuan and Qwen Token Plan Discounts

    Alibaba Cloud's 2026 Double 11 mega promotion runs from November 1 to 30. ECS cloud servers with 2 vCPUs, 2 GB of memory, and 3 Mbps bandwidth cost 99 yuan for one year, available to both new and existing users and eligible for renewal at the same price; lightweight application servers with 2 vCPUs, 2 GB of memory, and 200 Mbps peak bandwidth are available for 38 yuan for one year in a limited flash sale. On the AI side, the Qwen venue's Token Plan offers new customers discounts of up to 50% off, and the AI general savings plan covers more than 150 models including Qwen, Wan, GLM, and DeepSeek, with new customers getting up to 55% off. Individual spend-and-save gift packages are worth up to 360 yuan, enterprise packages up to 1,728 yuan, and newly certified enterprises can additionally receive AI gift packages worth up to 2,728 yuan.

    Same story · 20 reports

    #Industry news#Agents