24-Hour AI News: Millisecond Computing in Metro Areas Takes Place, Autonomous Driving and Multimodality Emerge

In the past 24 hours, the domestic focus was on "accelerating computing power and networks, implementing intelligent bodies and embodied intelligence, and new terminal AI products", while the foreign focus was on "autonomous driving overseas cooperation, platform AI compliance and financial market expectations". Domestic and foreign dynamics were basically balanced, covering multiple dimensions of policy, industry, products and security.

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

1. A notice was issued on October 16, proposing that by 2027, a metropolitan area millisecond computing network capability system with full coverage will be basically formed.
2. Key indicators: millisecond interconnection of computing centers (<1 millisecond), millisecond access to computing resources (<1 millisecond), and millisecond reachability of application terminals (network latency <10 milliseconds).
3. It will drive the application of high-speed optical transmission and all-optical cross-connection equipment of 400Gbps and above, and support the optimization of AI training and reasoning communication performance.

2. 2025 World Intelligent Connected Vehicle Conference opens in Beijing

1. The conference will open on October 16 and last for three days, with forums and side events focusing on six major themes: policy, technology, security, artificial intelligence, application, and data.
2. Release the interim results of the "vehicle-road-cloud integration" application pilot and initiate the construction of a national artificial intelligence automotive application pilot base.
3. The ministries and Beijing Municipal Government proposed to accelerate the deep integration of AI and transportation, and promote the multi-scenario application of high-level autonomous driving and international cooperation.

3. Zhou Hongyi: AI Enters a New Stage of “Intelligent Agent Development”

1. On October 17, it was proposed that AI will shift from "model training" to "application implementation", and intelligent agents will transform reasoning generation power into executable "digital employees".
2. Three factors for the rise of intelligent agents: reduced computing costs, open source of large models, and improved reasoning capabilities.
3. It is expected that the demand for computing power may increase by orders of magnitude after it becomes popular in the industry, so it is necessary to plan ahead.

4. AI Privacy and Security Draw Attention: The "Tripartite Collaboration Bridge" Between Endpoints, Cloud, and Users

1. On October 17, multiple parties called for strengthening authority and data governance at the terminal, application and user sides.
2. The notice shows that some AI applications have illegally collected personal information; security vulnerabilities in AI drawing tools have been exploited by foreign organizations to carry out attacks.
3. The "Mobile Agent Security Technology White Paper" proposes a confidential computing path of "intrinsic security" and end-cloud collaboration.

5. Yingmu Technology releases INMO GO3 AI smart glasses

1. Released on October 16, priced at 2,999 yuan, weighs about 53 grams, uses monochrome MicroLED and removable dual batteries.
2. It focuses on AR+AI ecology, two-way conversation real-time translation, AI onomatopoeia and other functions, and emphasizes all-weather battery life and privacy-proof design.
3. The industry chain disclosed that JBD supplies MicroLED, and offline channels and ecological platforms are promoted simultaneously.

6. Zhiyuan Robotics releases the Elf G2 and starts the first batch of commercial deliveries

1. On October 16-17, the company released the new generation of industrial-grade interactive embodied working robot G2, which has received orders worth hundreds of millions of yuan and completed the first batch of deliveries.
2. Positioning wheeled humanoid/embodied operations, emphasizing "cross-wrist force control arm", multimodal interaction and industrial-grade implementation.
3. It will be mass-produced in manufacturing scenarios such as automotive parts and consumer electronics, marking the accelerated commercialization of embodied intelligence.

7. Stellantis and Pony.ai sign an agreement to develop autonomous driving in Europe

1. The cooperation was announced on October 17, with priority given to the light commercial vehicle sector, targeting SAE L4 level.
2. The first batch of tests will be conducted in Luxembourg using the Peugeot e-Traveller, with plans to expand to multiple European cities starting in 2026.
3. Pony.ai's Luxembourg entity is responsible for European operations, emphasizing security, performance and compliance.

8. Meta strengthens control over AI chats with minors

1. On October 17, it stated that parents will be allowed to turn off one-on-one private chats between teenagers and AI characters, and can block specific AI characters.
2. The youth AI experience will be based on the PG-13 standard and will be launched in the US, UK, Canada, and Australia in phases.
3. Respond to previous criticisms of "flirtatious" AI chats and improve supervision and default protections.

9. UBS raises global stock market rating to "attractive"

1. Judgment given on October 17: The AI capital expenditure cycle brings about improved productivity expectations and revenue visibility.
2. We simultaneously upgraded the ratings of U.S. stocks and revised upward the year-end target for the S&P 500, believing that the AI-driven structural trend remains unchanged.
3. It suggests that concerns about valuations and bubbles still exist, but AI-related strategic cooperation has strengthened the sustainability of the economic boom.

10. Encord releases the world's largest open-source multimodal dataset and introduces the EBind method

1. Released on October 17: The dataset contains approximately 1 billion data pairs and 100 million data groups, covering text, images, video, audio and point clouds.
2. The accompanying EBind training method is said to be able to train a multimodal model with approximately 1.8 billion parameters in "hours" on a single GPU, and in some evaluations it can outperform larger models.
3. Advocate the use of data quality rather than "computing power" to promote the popularization of multimodal AI, and assist in the application of robots, autonomous driving and unmanned systems.

Frequently Asked Questions (Q&A)

Q: What are the hard indicators of “millisecond computing” in the metropolitan area?

A: Three core goals will be achieved by 2027: millisecond interconnection and millisecond access to computing centers (both <1 millisecond), one-way network latency from terminals to computing centers <10 milliseconds, and promotion of 400Gbps and above high-speed optical transmission and all-optical cross-deployment.

Q: What are the significant signals in this round of domestic AI applications?

A: The implementation of intelligent bodies and the commercialization of embodied intelligence are carried out simultaneously (Zhou Hongyi explained the shift to intelligent bodies and the mass delivery of Zhiyuan G2), superimposed with end-side AI devices (INMO GO3) and "millisecond computing" network speedup, forming a closed loop of "end-network-cloud-body".

Q: What are the new trends in overseas regulation of the use of generative AI by minors?

A: The platform strengthens parental controls and default protections (Meta allows parents to turn off private chats and PG-13 ratings), and collaborates with regulators in various countries to focus on risks such as inappropriate content, addiction, and data protection.

Q: Why are investment banks upgrading their stock market views related to AI at this time?

A: Institutions believe that AI capital expenditures and industry collaboration will improve corporate revenue visibility and productivity expectations, while the policy environment is also improving marginally. However, this also points to the risk of high valuations and bubbles, necessitating attention to fundamental realization and earnings expansion.