24-hour AI news: The "100 billion-level" model of central enterprises was released, and the OpenAI-Microsoft MOU was implemented

In the past 24 hours, industry-level large models, vertical models of government affairs and policies of ministries and commissions have been implemented in China; Overseas, it focuses on the ecological restructuring of large models, sovereign computing power investment, and the acceleration of government and enterprise-level AI applications, achieving balanced coverage of 6 articles at home and abroad, with a total of 12 articles.

1. China Logistics Group released a 278 billion parameter "flowing cloud" large model

1. The parameter scale reached 278 billion, positioning the industry-level general base and focusing on the full-stack scenario of the logistics supply chain.
2. It is jointly developed by China Logistics, China Telecom, China Mobile, iFLYTEK, and Huawei, emphasizing the collaboration of multi-party data and computing power.
3. Officially released on September 12, it is regarded as a milestone of the "industry model" led by central enterprises.

2. WeChat "Search" Grayscale launches AI search

1. The grayscale test was launched on September 12, adding "deep thinking", multi-modal questions (pictures/files, etc.) and conversational search.
2. According to reports, the ability is connected to models such as Hunyuan T1 and DeepSeek R1 to optimize the accuracy of Q&A search.
3. Strengthen "generative search" with entry-level capabilities, bringing a new round of iterative signals to the domestic search experience.

3. Ministry of Industry and Information Technology: Promote the deep empowerment of artificial intelligence to the entire pharmaceutical industry chain

1. On September 12, authoritative information was released, proposing to systematically promote "artificial intelligence + medicine" in R&D, production, circulation, supervision and other links.
2. Emphasize technologies such as privacy computing and trusted data space to escort data rights and interests.
3. Clearly emphasize the idea of paying equal attention to industrialization and compliance, and expand the depth of "AI + medicine".

4. The country's first "administrative reconsideration" vertical model was released

1. The Beijing Municipal Bureau of Justice released the "administrative reconsideration" model during the CIFTIS, with a nominal output accuracy rate of more than 95%.
2. Respond to the rapid growth of caseload: In 2024, there will be 749,600 new cases nationwide, a year-on-year increase of 94.7%.
3. Position it as a paradigm model for intelligent case handling of government affairs, and explore public data training and compliance paths.

5. AsiaInfo and Tsinghua AIR released the world's first AI-RAN business white paper

1. On September 12, the "AI-RAN Business Outlook White Paper for Vertical Industries" was released and included in arXiv.
2. System modeling verifies the commercial feasibility of "Private AI-RAN" and proposes a path for enterprise network modernization.
3. The joint parties include Intel, Tsinghua AIR, Hong Kong Polytechnic AI Research Institute, IEEE, etc., and the industry-academic-standard linkage.

6. The National Development and Reform Commission and other departments released "artificial intelligence" support signals

1. According to multi-channel information on the same day, the National Development and Reform Commission proposed to increase financial and financial support in the field of AI.
2. Promote the pilot of 10,000 gigabit optical network and intelligent computing infrastructure at the local level to serve the implementation of AI industrialization.
3. The "investment-construction-use" collaboration on the policy side is expected to accelerate the two-wheel drive of computing power and applications.

7. OpenAI and Microsoft signed a non-binding MOU, entering a new stage of cooperation

1. The official joint statement on September 11 confirmed the signing of the MOU and is promoting a binding final agreement.
2. The two sides reaffirmed that they will provide the best AI tools on the premise of security, which will pave the way for governance and business structure adjustments.
3. It echoes the process of Optimizing OpenAI's governance and equity structure, affecting the reshaping of the ecological pattern.

8. According to media reports: OpenAI and Nvidia plan to announce data center investment in the UK

1. On September 12, a number of media said that the two will work with the UK to promote the construction of "sovereign AI" infrastructure, which may reach billions of dollars.
2. The project belongs to the global computing power map layout with the large computing power plans such as "Stargate" in the United States.
3. The incident is still in the media reporting stage, and the details have yet to be officially confirmed, reflecting the deep linkage between the United States and the United Kingdom in AI infrastructure.

9. The British government announced the results of the AI coding assistant pilot

1. On September 12, the British government announced: more than 1,000 technicians and 50 departments participated in the pilot.
2. The average per capita savings is nearly 1 hour/day, which is equivalent to about 28 working days per person per year.
3. The government level clarifies the quantitative basis for using AI tools to drive "public sector efficiency".

10. Anthropic launches memory function for Claude Enterprise/Team Edition

1. On September 12, the memory function will be opened to Team and Enterprise users, supporting cross-session context retention.
2. Focus on the personalization and continuous collaboration of enterprise scenarios, taking into account compliance and controllable switches.
3. Combine with enterprise knowledge base/security strategy to improve the practicability of "long-term task" AI agents.

11. Meta launches a large model-driven NPC tool for Worlds developers

1. Officially launched on September 12, developers can create customizable NPCs that respond to player voices.
2. Integrate multiple generative AI tools to accelerate resource and script generation and lower the threshold for content production.
3. Another progress in the commercialization of "generative AI + interactive content".

12. The key provisions of the EU Data Law come into effect today

1. From September 12, obligations such as data portability and cloud service migration will take effect to reduce the risk of cloud lock-in.
2. Cover personal and non-personal data and strengthen the B2B/B2G data sharing framework.
3. It has a far-reaching impact on AI training data acquisition, enterprise data governance and cloud computing power procurement strategies.

Frequently Asked Questions (Q&A)

Q: What exactly does the MOU between OpenAI and Microsoft mean?

A: The two sides have entered the "non-binding->-restrictive" transition period and reaffirmed their long-term cooperation and security commitments. For the ecosystem, it is conducive to clarifying the coordination and division of labor between future products, computing power and business boundaries.

Q: What is the most hard-core model progress in China today?

A: China Logistics Group released a 278 billion parameter "flowing cloud" large model, and central enterprises and operators cooperated with leading manufacturers to position the industry-level base and highlight the collaborative path of "industry + computing power + data".

Q: What quantitative results do the UK government's AI coding pilot provide?

A: Covering 50 government departments and 1,000+ technical personnel, the average saving per person is nearly 1 hour/day, which is equivalent to an efficiency improvement of about 28 working days per year, which has demonstration significance at the policy and procurement level.

Q: What are the direct impacts of the EU Data Law on AI companies?

A: Data portability and cloud migration obligations take effect, and supplier lock-in costs are reduced; It is beneficial to the compliance of cloud strategies and data sharing, and indirectly improves the data and computing power supply elasticity of AI training/inference.