Goldman Sachs Chief Investment Officer Looks Ahead to 2024: Hybrid AI Takes Center Stage

In 2024, Goldman Sachs' Chief Investment Officer predicts that the era of hybrid artificial intelligence will dawn, with applications rising to prominence as a key factor. Under the constraints of foundational models, businesses will focus on building small-scale networks. 2024 will see the rise of third-party applications, forming new application and tool layers around foundational models. Emphasis will be placed on secure collaboration and risk management. Capital will flow towards the application and tool layers, propelling the continuous development of the artificial intelligence field.

QuestMobile: Doubao and DeepSeek Exceed Other AI Apps in Monthly Uses per User

QuestMobile released the China Mobile Internet Autumn Report on November 3, stating that the domestic battle for AI application entry points is nearing its end, with competition focus shifting from user scale expansion to high-frequency interaction and user operations. Application entry points with independent service closed loops, as well as system-level entry points that can enable service delivery, are gradually gaining an advantage. In September 2026, Doubao and DeepSeek reached 72.0 and 65.9 average monthly uses per user, respectively, making high-frequency usage a key indicator.

8GB VRAM Can Also Handle 128K Ultra-Long Context! Open-Source Local Large Model Tool Strata 0.2.10 Released

Open-source local AI tool Strata releases version 0.2.10, specially optimized for consumer-grade GPUs. This update focuses on lowering the VRAM threshold, improving inference speed, and optimizing multi-GPU collaboration, making it easier for ordinary computers to run hundred-billion-parameter large models. The highlight is a significant reduction in VRAM usage: 8GB graphics cards can also support a 128K ultra-long context, whereas previously enabling contexts above 64K with 8GB VRAM was often restricted. Overall, it makes local large model deployment more accessible and efficient.

Alibaba Open-Sources 470 Qwen Models; Joseph Tsai: Open Source Is the Key to AI Adoption

The 2026 World Internet Conference opened in Wuzhen on November 3. Joseph Tsai disclosed that Alibaba Qwen has open-sourced more than 470 models, with global downloads exceeding 4 billion and more than 330,000 derivative models. He believes that as AI demand rises and computing power is in short supply, only open source can popularize AI. Open source can reduce R&D costs for SMEs and is especially critical for developing countries, reflecting the open-source philosophy of Chinese companies 'opening up like top universities'.

Tencent Hunyuan releases an AI Scientist: it trains proteins, builds robotic hands, and even improves next-generation AI

Tencent Hunyuan's Hyra 2.0 enables AI to autonomously build reusable scientific models, algorithms, and tools. A three-tier Agent (Overall/Mission/Worker) manages coordination, planning, and experiments with persistent states for parallel projects. Conclusions need code and logs, verified by independent Agents; insights become reusable Skills, spanning protein to chip research.....

From Prohibiting Military Use to Winning Over the Pentagon: OpenAI Hires Former Army Secretary as Advisor

On November 3, OpenAI announced that it had hired Daniel Driscoll, who just resigned as U.S. Secretary of the Army in August, as Senior Advisor for National Security and Modernization, to drive the implementation of AI in national security and cybersecurity. In a statement on the 2nd, the company said he will collaborate with government, academia, and industry to advance modernization reforms and advise public and private partners, leveraging AI to strengthen national security capabilities. Driscoll is 38 years old.