NAll news

Wednesday916 items

  1. Shao MengAnalysis
    AI score51

    Playwright officially clarifies browser security: MCP and CLI use stable Chrome, the test runner pins versions and disables the sandbox by default

    Playwright officially clarifies that no Playwright package bundles a browser, so there is no need to delete it because of vulnerability rumors. Agents using Playwright MCP or the CLI invoke the stable Chrome or Edge on your machine, which receives security updates normally and has the sandbox enabled by default; whereas the Chrome for Testing downloaded by npx playwright install has a pinned version and the sandbox disabled by default, making it suitable only for end-to-end testing of trusted applications. When calling chromium.launch() directly, the default is likewise a pinned version with the sandbox disabled; if the code will access uncontrolled pages, switch to chromium.launch({ channel: 'chrome', chromiumSandbox: true }) and keep Chrome and Edge up to date.

    Playwright and browser security reminder. We saw a post suggesting that you delete Playwright because it bundles browsers with CVEs. That is a misconception. Do not delete Playwright🙃. Here is how it works and what you should do. No Playwright package bundles a browser. Agents on Playwright MCP or CLI run your stable Chrome, which gets Chrome's security updates, with the sandbox on. 🤖 If you use Playwright MCP or the Playwright CLI They use the stable Chrome / Edge on your machine, so it gets its normal updates. No config edit is needed. 🧪 If you use the test runner When you run npx playwright install, Playwright downloads Chrome for Testing, pinned to the build we tested that release against. The pin is on purpose. Your tests run on the same browser on every laptop and CI machine, and a browser update cannot suddenly break your suite. You choose when to move by upgrading Playwright. New releases usually pin a build ahead of stable Chrome, so you see breakage before your users do. We recommend this for full e2e test suites on your own trusted app. This build can miss Chrome security fixes until the next Playwright release, and by default it runs with the sandbox off. Do not use it on the hostile web. 📚 If you use the library API chromium.launch() has the same defaults: the pinned build with the sandbox off. If your code visits pages you don't control (scrapers, agent scripts), max out protection: chromium.launch({ channel: 'chrome', chromiumSandbox: true }) 🔒 Always keep Chrome / Edge updated. Some bugs can escape a sandbox. Happy testing!

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

  2. AYi AI Notes
    AI score32

    Flatkey Uses Seedance to Generate a 15-Second Multi-Character Dance Video

    The author called the Seedance model on Flatkey and generated a 15-second dance video featuring multiple characters in the same frame, with Sam Altman and Dario Amodei dancing side by side; the actual cost was under $0.1. The author shared three Prompt pitfall-avoidance parameters: use absolute positioning to lock down character features to prevent face swapping, unify the choreography rhythm to prevent limbs from melting, use a 35mm wide-angle cinematic lens to keep the characters on the same focal plane, and add a no face morphing constraint. Flatkey's everyday pricing is at least 20% lower than official pricing, failed requests cost $0, and it provides access to 100+ official models.

    English text is being prepared.

    #Tutorials / practice#Video#Agents

  3. TMTPost · AGI
    AI score12

    Edge AI Daily Morning Brief (November 4): Anthropic's $895.4 Billion Compute Commitment and GPT-6.2…

    On November 2, Anthropic filed a revised S-1, quantifying for the first time approximately $895.4 billion in long-term compute and infrastructure commitments, about seven times its $132 billion in cash and short-term investments, with IPO discussions valuing the company at up to roughly $2 trillion. During the same period, some OpenAI Codex users spotted an undisclosed identifier "gpt-6.2-astra" on the usage statistics page, accounting for as much as 19.8% of a single account's daily requests, which the company has not acknowledged. Meta, meanwhile, launched its first investment-grade U.S. dollar bond sale in six months, with the 30-year tranche priced at about 7.8%, using the proceeds to fill the AI infrastructure funding gap.

    #Industry news#OpenAI#Anthropic

  4. TMTPost · AGIAnalysis
    AI score59

    Hands-on test of Today AI and Ant Lingguang: Which is closer to a Chinese version of Muse

    Guangzhui Intelligence conducted hands-on tests of Today AI, built by Qi Junyuan, former head of Doubao's PC team, and Ant Lingguang, upgraded to version 2.0. The former is positioned as a personal assistant that builds user profiles by reading local files, photo albums and text messages and proactively pushes suggestions; the latter is positioned as a life butler that analyzes spending habits based on Alipay deduction records. In the tests, Today AI was more proactive on phone bill discounts, web price comparisons and travel suggestions, while Lingguang was more reliable on scheduled tasks and organizing meeting notes, but neither could invoke apps such as Taobao, JD.com or 12306, and cross-app data retrieval remains constrained by super-app walled gardens.

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

  5. TMTPost · AGI
    AI score36

    AI Games Enter the "Hongguo Moment": Claude Opus 5.5 Boosts Game Programming, Why Are Tencent and miHoYo Unfazed?

    Anthropic releases Claude Opus 5.5, focusing on enhancing game programming capabilities and driving large models to reshape game development. The share of games on Steam tagged with AI rose from 1.2% in 2023 to 30.8% by mid-2026, and Musk also founded the XAI artificial intelligence game studio, planning to release a native AI game based on Grok before the end of 2026. WeChat Mini Games and Douyin Mini Games, meanwhile, upgraded their rights-protection mechanisms to address copyright issues caused by decompilation and cloning.

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

  6. TMTPost · AGI
    AI score29

    AI drama "Yu Tang" female lead accused of resembling creator; creative team announces full redo of Season 3

    After the launch of Season 3 of the AI drama "Yu Tang," viewers questioned that the modeling of female lead Ke Miantang was too similar to creator Sanqi herself, and were dissatisfied over the appearance in the "wrap" easter egg of a character with suspected personal traits of the creator. Sanqi posted a 2,282-character apology letter denying face-merging, but admitted problems with the plot arrangement, easter egg design, and personal responses, and announced that the team would rewrite and remake Season 3, while she would withdraw from public online operations. The series had previously gained 1.94 million Hongguo favorites and a platform peak popularity of 46.53 million.

    #English text is being prepared.#Video

  7. TMTPost · AGI
    AI score38

    AI digital employees enter the headcount, and the historical mission of interns may be over

    JD.com is running a partial pilot within its finance CFO system, incorporating AI digital employees into the P3 job grade and setting up independent headcount; the work performance and direct supervisor evaluations of some Agents have already surpassed those of interns. An August 2026 study by the Stanford Digital Economy Lab found that in industries susceptible to AI, the employment level of young people aged 22 to 25 was about 19% lower than the original trend, and the impact was mainly reflected in companies reducing the hiring of new employees. In a January 2026 experiment by Anthropic, 52 junior software engineers scored an average of 50% with AI assistance, while the handwritten-code group scored 67%.

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

  8. TMTPost · AGIResearch
    AI score56

    Shanghai AI Laboratory Releases PrivacyTrace Benchmark to Quantify Privacy Leakage in Agent Tool-Calling Traces

    Shanghai AI Laboratory, together with the University of Electronic Science and Technology of China and the University of Sydney, released the study "How Much Can Tool Calling Expose You?", building the PrivacyTrace benchmark. Using 1,000 synthetic users and 25 real MCP servers, they produced 4,000 traces comprising 67,950 tool-call records in total, and had four models—DeepSeek-V4-Flash, GPT-5.5, Gemini-3-Flash, and MiniMax-M3—infer 17 user attributes from the traces alone. The results show that the attack success rate rose from 17.97% for blind guessing to 26.25% for single-domain traces, and reached 32.77% after chaining across four domains, with nationality (57.73%) and religious belief (50.03%) being the most severely leaked. Ablation experiments show that tool parameters contribute the most; adding parameters alone raised the rate from 14.53% to 27.71%. Based on this, the paper proposes three defense directions—obfuscation, minimization, and severing cross-service correlations—but states that they have not yet been experimentally validated.

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

  9. English text is being prepared.Model releases
    AI score58

    Baidu Discloses Apollo World Model for the First Time, Virtual Road Testing Covers 100,000 Kilometers per Day

    Baidu Apollo has disclosed for the first time Apollo World, an omni-modal world model for physical AI, and proposed a "2+3" evaluation criteria for world models. The model's core capabilities are agent-driven operation, omni-modal perception, and trip closed-loop, and it can generate camera images and LiDAR point clouds in real time for the driving system to perceive and make decisions and to update the traffic environment. Currently, Apollo World can complete about 100,000 kilometers of virtual road testing per day and has been used in the global testing and validation of Apollo Go.

    Same story · 8 reports

    #Model releases#Embodied AI#English text is being prepared.

  10. English text is being prepared.
    AI score25

    Seyond Joins Forces with Dell and NVIDIA to Showcase Real-Time Spatial Digital Twin, Accelerating the Integration of LiDAR and Physical AI

    LiDAR company Seyond, together with Dell and NVIDIA, showcased a LiDAR-based real-time operational digital twin system at the 2026 World Smart City Expo. The system uses a Dell system equipped with NVIDIA DGX Spark, combining a 20-core Arm Grace CPU with an NVIDIA Blackwell GPU. The solution integrates a visual AI agent built on the NVIDIA Metropolis VSS Blueprint and NVIDIA Cosmos, and can be used for scenarios such as pedestrian flow analysis and space utilization monitoring. Seyond also announced a partnership with AMSYS to combine its LiDAR perception technology with the AMSYS Fusion Core Perception Platform, providing spatial intelligence solutions for large venues and public spaces.

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

  11. English text is being prepared.
    AI score35

    Linker Vision Releases Linker AI Nexa, Integrating 3DGS and AI Agent to Empower Urban Governance

    Linker Vision unveiled the Linker AI Nexa platform at the 2026 Smart City Expo World Congress in Barcelona, integrating 3DGS, digital twins and AI Agent for urban public safety, disaster response and environmental governance, while incorporating technologies such as NVIDIA Metropolis, Cosmos, Omniverse and Nemotron. In a Taipei pilot, 3DGS scanning increased a single person's daily tree survey volume from about 100 trees to more than 1,000 trees, and can simulate tree canopy growth and shading changes over the next 15 years in about 10 minutes. Simulations show that after optimizing the planting plan, tree shade coverage is expected to increase from 15.7% to more than 50% over the next 15 years.

    #Product updates#Agents#Multimodal

  12. English text is being prepared.
    AI score31

    MR-assisted surgical system HipInsight expands implant support, compatible with over 35 hip prosthesis systems

    Surgical Planning Associates' (SPA) MR-assisted surgical system HipInsight has added five orthopedic implant systems, bringing its preoperative 3D planning database to support more than 35 implant systems, covering primary and revision total hip arthroplasty procedures. The newly added manufacturers include Enovis, Maxx Orthopedics, MicroPort Orthopedics, Smith+Nephew, and Zimmer Biomet. The system is based on Microsoft HoloLens 2 and can overlay 3D images of the patient's pelvis and other anatomical structures onto the actual surgical field during surgery to assist with implant positioning.

    #Product updates#Multimodal

  13. English text is being prepared.
    AI score47

    SenseTime Releases SenseNova 6.8 Flash Official Version, Supporting Dynamic 3D Blender Design

    SenseTime has launched the official version of SenseNova 6.8 Flash, adding AI Native Design capabilities that support dynamic 3D Blender project generation and iteration. This version focuses on agents' long-term memory and continuous task collaboration across sessions. The Agent can first generate an initial version of a scene, then make multiple rounds of revisions based on rendering results and user feedback, and finally output a Blender project that can continue to be edited.

    Same story · 5 reports

    #Model releases#Agents#Multimodal

  14. English text is being prepared.
    AI score56

    Xiaomi's humanoid robot works continuously for 8 hours in a factory, completes over 50 tasks in youth apartment

    Xiaomi disclosed the latest progress of its robotics business: its humanoid robot completed an 8-hour continuous operation challenge at the center console side cover sequencing station in an automobile factory, with the ability to autonomously correct operational anomalies and coordinate with the production line. An Agent system built on the Xiaomi-Robotics-1 embodied foundation model completed long-horizon task validation in Xiaomi youth apartments, operating continuously for over 30 minutes, spanning 3 rooms, and covering more than 50 tasks. Xiaomi-Robotics-1 is pretrained on 100,000 hours of real-world manipulation trajectories and combines post-training data to optimize manipulation capabilities.

    Same story · 5 reports

    #Industry news#Embodied AI#Agents

  15. Shanghai Observer
    AI score27

    2026 World Internet Conference Wuzhen Summit: Open Source and Openness Drive Inclusive AI

    The 2026 World Internet Conference Wuzhen Summit opened on November 3 in Wuzhen, Zhejiang, under the theme "Open Source, Openness, Co-construction and Sharing—Jointly Building a Community with a Shared Future in Cyberspace." Alibaba's Joseph Tsai introduced that Alibaba's Qwen has open-sourced more than 470 models, with cumulative global downloads exceeding 4 billion and more than 330,000 derivative models, ranking first in the world; a nonprofit organization in Uganda has developed a local language model covering 68 African languages based on the Qwen open-source model. UN Under-Secretary-General Li Junhua pointed out that about 2.2 billion people worldwide still lack internet access.

    #Industry news#Open-source ecosystem#Agents

  16. Semiconductor Industry Observer
    AI score70

    AMD's Q3 2026 revenue reached $13.4 billion, with data center revenue share rising to 64%

    AMD announced its Q3 2026 earnings: revenue of $13.4 billion, up 45% year-over-year; data center business revenue of $8.5 billion, up 97% year-over-year, accounting for 64% of total revenue, up from 47% in the same period last year. Lisa Su stated that in 2027 the data center business will account for more than 80% of the company's total revenue, and in Q4 data center AI business revenue will surpass server CPU business revenue for the first time. EPYC Venice has begun shipping, MI455X has started volume shipments, and customers including Anthropic, Meta, Microsoft, OpenAI and Oracle completed their first Helios rack deployments in Q3.

    Same story · 6 reports

    #Industry news#English text is being prepared.

  17. Semiconductor Industry Panorama
    AI score45

    Morgan Stanley CEO: AI Infrastructure Will Cost $10 Trillion to $15 Trillion Over the Next 10 Years

    Ted Pick, Chairman and CEO of Morgan Stanley, predicts that capital expenditure in the AI sector will need to reach $10 trillion to $15 trillion over the next 10 years, driving GDP growth of 50 to 70 basis points. He noted that the key lies in converting investment into actual productivity gains rather than simply piling up computing power; power supply is becoming a hard constraint limiting data center expansion.

    Same story · 5 reports

    #Industry news#English text is being prepared.

  18. Semiconductor Industry Observer
    AI score63

    onsemi Q3 Data Center Revenue Up 175% YoY, Acquires Ferric for Up to $775 Million to Bolster AI Power Supply

    onsemi reported third-quarter revenue of $1.722 billion and adjusted earnings per share of $0.90, both above analyst expectations, with data center revenue of $180 million, up 39% quarter over quarter and up 175% year over year. The company announced the same day that it will acquire integrated voltage regulator (IVR) maker Ferric for $600 million in cash, plus up to an additional $175 million contingent on revenue milestones, for a total of up to $775 million, while raising its 2030 AI data center revenue target to more than $3 billion.

    Same story · 2 reports

    #Industry news#English text is being prepared.