Back to AI News Briefing
24-Hour AI News: Metro Area "Millisecond Computing" Implementation and System-Level Copilot Upgrade

24-Hour AI News: Metro Area "Millisecond Computing" Implementation and System-Level Copilot Upgrade

AI News Briefing Admin 64 views

In the past 24 hours, domestic ministries and commissions issued documents to promote "millisecond computing", the commercial delivery of embodied robots, the release of AI intellectual property group standards, and the opening of the Intelligent Connected Vehicle Conference; overseas, Microsoft strengthened Windows 11 Copilot, Anthropic released Claude Haiku 4.5, Apple launched the M5 chip, and British regulators accelerated the pilot of AI medical tools and revealed the trend of global public opinion.

1. MIIT launches special action on "Millisecond Computing" in metropolitan areas

  1. An official notice was released on October 16, focusing on building a high-speed, large-capacity, and low-latency metropolitan computing network.
  2. Goal by 2027: interconnection latency of computing centers is less than 1 millisecond, and network latency of application terminals is less than 10 milliseconds.
  3. It is proposed that the coverage rate of the "1 millisecond delay circle" should be no less than 70%, and the full-area coverage capability system should be promoted in stages.

2. Zhiyuan Robotics launches the "Elf G2", announces orders worth hundreds of millions of yuan and begins first batch of deliveries

  1. The product will be released online and offline on October 16th. It is positioned as an industrial-grade interactive embodied working robot and is suitable for multiple scenarios such as industry, logistics, and guided tours.
  2. It was disclosed that the company has received orders worth hundreds of millions of RMB; and has started the first batch of commercial deliveries under a procurement contract worth over 100 million RMB with Joyson Electronics.
  3. The product continues the "acquisition, training and promotion in one" route, emphasizing rapid learning and multimodal interaction, and aims at large-scale industrial implementation.

3. Beijing investigates and punishes the first case of misusing AI technology to publish false advertisements

  1. On October 16, the Beijing Municipal Market Supervision Department announced: Investigate and punish illegal advertisements that use AI to impersonate public figures (including hosts) to promote health products.
  2. Emphasize the supervision of and accountability for the generation of synthetic content (deep fake) advertisements.
  3. Release the law enforcement signal that “AI advertising must be labeled and evidence retained in compliance with regulations.”

IV. China's first "Generative AI Intellectual Property Guidelines" group standard was released and implemented in Haidian

  1. Released on October 16, standard number T/CECC 42-2025, has been implemented since October 10.
  2. Covering data compliance, patents and trade secrets, model review and complaint mechanism, trace evidence collection, etc.
  3. Clarify compliance boundaries for model developers, service providers and users, and improve dispute resolution efficiency.

5. The World Intelligent Connected Vehicle Conference opened in Beijing, with intelligent driving rules and global coordination becoming the focus.

  1. The conference will open on October 16 in Yizhuang, Beijing, and will focus on international docking of autonomous driving rules, vehicle-road collaboration and standards.
  2. The industry is concerned about L3/L4 implementation, data compliance, algorithm security, and responsibility determination.
  3. Release the signal of "rules first + industry collaboration" to promote the clarification of the commercialization path of intelligent connected vehicles.

6. Microsoft launches a new round of AI upgrades for Windows 11, systematically expanding Copilot capabilities

  1. Released on October 16: "Hey Copilot" voice wake-up, Copilot Vision fully open for testing, and desktop "Copilot Actions" to perform real-world tasks (such as ordering food reservations).
  2. Launch "Gaming Copilot" for gaming scenarios and expand the available feature set of the Insider channel.
  3. The strategy is to achieve system-level integration of “AI as the operating layer” to consolidate the competitiveness of the desktop ecosystem.

7. Anthropic Releases Claude Haiku 4.5, Bringing Mid-Range Performance to Small Models and Significant Cost Reduction

  1. Released on October 15: Approaching or matching Sonnet 4 levels in coding and computer operation tasks.
  2. The cost is about one-third of Sonnet 4, and the cost of higher-order models is about one-fifteenth, with faster inference speed.
  3. An economical option for large-scale enterprise deployment and wide-area business process automation.

8. Apple releases the M5 chip, significantly improving edge AI computing power and bandwidth

  1. October 15th press release: Compared with M4, AI-related peak GPU computing power has increased by more than 4 times.
  2. Unified memory bandwidth is increased to 153GB/s; the neural network engine has 16 cores, enhancing the ability to run large models on the device.
  3. Upgraded AI localization experience for MacBook Pro, iPad Pro and Apple Vision Pro.

9. UK MHRA launches Phase II of AI Airlock, with seven AI medical tools entering pilot programs

  1. Multiple reports on October 16: Seven AI technologies were selected to enter the second phase of Airlock to accelerate their verification in NHS diagnosis and care.
  2. Continuing the reforms implemented in early October: the average time for clinical trial approval was reduced from 91 days to 41 days (supporting processes and AI tools working together).
  3. The regulatory approach emphasizes "AI-assisted review and manual decision-making" to speed up the implementation of innovation while ensuring safety.

10. Global Poll: The American Public is More Cautious About AI, with Differences in Gender and Education

  1. A cross-national survey released on October 15 showed that American respondents were more concerned about AI; men and highly educated groups around the world were more accepting of AI.
  2. Trust in AI regulation in China and the United States is generally low, reflecting a cognitive gap in cross-border governance.
  3. Public opinion will affect corporate AI deployment strategies and policy communication paths.

Frequently Asked Questions (Q&A)

Q: What will be the key indicators of the “Millisecond Computing” project by 2027?

A: The interconnection delay of the computing power center is less than 1 millisecond, and the network delay of the application terminal is less than 10 milliseconds. The coverage rate of the "1 millisecond delay circle" is not less than 70%, and it is constructed in different regions and stages.

Q: What is the biggest difference between Claude Haiku 4.5 and Sonnet 4?

A: The performance is similar in coding/computer operation tasks, but Haiku 4.5 costs about one-third and is faster, making it suitable for large-scale enterprise deployments.

Q: Does Apple's M5 significantly improve AI workloads?

A: Yes. Compared to the M4, the AI-related peak GPU computing power is increased by more than 4 times, the unified memory bandwidth reaches 153GB/s, and it is equipped with a 16-core neural network engine to enhance the inference of large models on the edge.

Q: What does the second phase of MHRA’s AI Airlock mean?

A: Seven AI medical tools have entered regulatory "sandbox" pilot programs; combined with existing reforms, this has accelerated review and approval processes (the average time has been reduced from 91 days to 41 days), ensuring safety and transparency in a human-machine collaborative model.

Q: What chain reactions will Beijing’s first AI false advertising case bring?

A: Advertisers need to strengthen "compliance identification + evidence retention", the review and responsibility of platforms and producers will be improved simultaneously, and the commercial dissemination of deep fake content will face higher law enforcement intensity.

Millisecond Calculation Special Metropolitan Area Computing Network 1 millisecond delay loop Low-latency computing network 2027 Computing Network Goal Computing center interconnection latency Application terminal network delay Zhiyuan Robot Elf G2 Commercial delivery of embodied robots Industrial-grade interactive robots Hundreds of millions of yuan in robot orders Joyson Electronics Procurement Cooperation Integrated recruitment, training and promotion route Beijing AI false advertising case Deep fake advertising regulation AI Advertising Compliance Label Intellectual Property Guidelines for Generative AI Group standard TCECC42_2025 Haidian releases AIIP standards Data compliance and trace evidence collection World Intelligent Connected Vehicle Conference L3_L4 landing path Vehicle-Road Collaboration and Algorithm Safety Autonomous driving liability determination Microsoft Windows 11 Copilot upgrade HeyCopilot voice wake-up CopilotVision is now fully open CopilotActions desktop commands GamingCopilot Game Assistant AnthropicClaudeHaiku4_5 Haiku vs. Sonnet4 Small models reduce costs and increase efficiency Enterprise-level process automation Apple M5 chip released M5AI computing power increased by 4 times Unified memory bandwidth 153GB_s 16-core neural network engine Apple's large-scale model inference MacBookPro_iPadPro_VisionPro UK MHRAAI Airlock Phase II 7 AI medical tool pilot projects NHS diagnostic AI validation The approval time was reduced from 91 days to 41 days Human-machine collaborative supervision path Global AI Poll The American public is more cautious Gender and educational differences Low trust in AI regulation between China and the US Adjusting enterprise AI deployment strategies AI News on October 16

Recommended Tools

More