AI Daily: Tencent Cloud Launches CodeBuddy NPC; Beijing Introduces Ten Policies for Intelligent Agents; Samsung's Galaxy Glasses Unveiled

Welcome to the "AI Daily" section! This is your guide to exploring the world of artificial intelligence every day. Every day, we present you with the latest content in the AI field, focusing on developers to help you understand technical trends and innovative AI product applications.

Hot AI products Click to learn more: https://app.aibase.com/zh

1. Samsung puts Gemini on its face: 50g Galaxy Glasses unveiled, no screen but equipped with all-day AI eyes

Samsung launched its first AI smart glasses, Galaxy Glasses, powered by Google's Gemini model, featuring a slim design and hands-free interaction, offering an all-day visual intelligent assistant experience.

【AiBase Summary:】

👓 Galaxy Glasses features a slim design, weighing only 50 grams, without a traditional AR display, capturing real-world images through a camera and analyzing them using the Gemini model.

🤖 Integrated with Google's multi-modal large model Gemini, it supports scenarios such as office work, travel, and communication, such as automatically organizing whiteboard text, voice navigation, and real-time translation.

🔋 Battery life lasts up to 9 hours, and with a charging case, it can be extended to 7 full charges, seamlessly integrating with the Samsung Galaxy ecosystem for balanced task distribution between devices.

2. Xiaohongshu open sources BigMac: Liberating memory and speed from the trade-off in multi-modal training

The article introduces Xiaohongshu's open-sourced BigMac, a new paradigm for multi-modal pipeline parallel training, solving the conflict between memory and speed in multi-modal model training. BigMac optimizes scheduling strategies through a quasi-dependency-safe nested pipeline design, improving training efficiency and verifying its effectiveness in practical applications.

【AiBase Summary:】

🧠 BigMac solves the conflict between memory and speed in multi-modal training through a quasi-dependency-safe nested pipeline design.

⚙️ Provides global scheduling tables, seamless interfaces, and toolchains to improve system integration and performance troubleshooting efficiency.

📈 Practical tests show that BigMac outperforms existing solutions in both memory and speed, suitable for complex multi-modal scenarios.

3. Tencent Cloud launches CodeBuddy NPC: From code assistant to end-to-end self-developed intelligent agent

Tencent Cloud launched CodeBuddy NPC, marking the entry of enterprise-level AI R&D ecology into a stage of deep engineering collaboration, significantly reducing the cost of large model applications and driving the comprehensive intelligent reconstruction of cloud-native R&D workflows.

【AiBase Summary:】

🧠 CodeBuddy NPC is a new cloud-based R&D intelligent agent launched by Tencent Cloud, deeply relying on the CNB R&D platform to achieve end-to-end self-development.

💻 Unlike traditional local AI coding tools, CodeBuddy NPC can independently complete the entire workflow, including solution planning, coding development, submitting PRs, and automated testing.

📊 By optimizing prompts, tool calls, and cache hit rates, CodeBuddy NPC reduces the first-round Token consumption by over 90% and supports flexible model switching based on task difficulty.

4. Beijing introduces ten policies for intelligent agents: From orchestration layer engineering to one-person companies, an Agent economy blueprint is being laid out

Beijing issued the "Several Measures on Accelerating the Development of Intelligent Agents," incorporating multiple technical terms into official documents, covering Agentic AI, Harness Engineering, AI OS, etc., to promote the development of the agent economy. The policy focuses on enhancing the ability to complete large model tasks, optimizing system architecture, encouraging innovation and entrepreneurship, building a Token economy system, and strengthening security governance.

【AiBase Summary:】

🧠 The grand blueprint of the agent economy has now received a clear outline at the policy level.

🛠️ Orchestration layer engineering requires optimizing context engineering, task persistence, and multi-agent collaboration to build a common foundation.

💼 Supports new models of innovation and entrepreneurship like OPC, lowering the minimum viable scale for enterprises.

5. China's First AIGC Story Film with a Network Drama License "Qitan" Launched on iQIYI

The article introduces the domestic first AIGC network story film "Qitan: Paper Blade Crossing the Wasteland" that has obtained the "Network Drama Film Distribution License" and is now available on iQIYI. Produced jointly by Daming Innovation and iQIYI, the film uses AI to complete the entire process from concept to production. It tells the story of the amnesiac paper master Bai Xiaoman and the half-paper person Hu Burui searching for memories in the gap between the Yin and Yang worlds, showcasing the aesthetics of Chinese strange tales. In production, the film systematically tested the AIGC workflow for long-form storytelling and deeply explored classical texts like "Shanhaijing," "Zi Buyu," and "Liaozhai," creating a unique visual world.

【AiBase Summary:】

🎬 The first AIGC network story film in China with a "Network Drama Film Distribution License" has been launched.

🧠 AI technology was used throughout the entire creative to production process, showcasing the aesthetics of Chinese strange tales.

💡 The production team deeply explored traditional strange tale classics to build a unique visual world.

6. DeepSeek Liang Wenfeng Talks About Open Source: Only Make 6 Times Profit Without Overtime, Want to Make 100 Times Profit Will Be Restricted by Open Source

DeepSeek founder Liang Wenfeng elaborated on the underlying logic of the open-source strategy during an investor exchange session, emphasizing that open source is an inevitable choice and analyzed the relationship between open source and commercial profit from a financial perspective. He pointed out that open source does not affect six times profit, but will restrict hundred times profit, while also believing that medium size is the company's unique advantage.

【AiBase Summary:】

✅ Liang Wenfeng clearly stated that he will definitely open source, even the strongest model will be open sourced.

💰 Explains the relationship between open source and commercial profit from a financial perspective, open source does not affect six times profit but restricts hundred times profit.

🌱 Positions the current scale as the sweet spot, believes that medium size is a unique advantage, small startups are too small, and big companies are too large to organize.

7. Gaode Releases ABot Embodied System Full Stack Upgrade, Unveils Five Core Models

Gaode released the full stack upgrade of the ABot embodied system, unveiling five core models, marking the transition of embodied intelligence technology to a highly collaborative and integrated new stage.

【AiBase Summary:】

🤖 ABot-N1 serves as a general navigation base, using a fast-slow dual-system architecture, balancing long-range logical reasoning and real-time control.

🕹️ ABot-M0.5 maps the "hands", achieving real-time alignment of eyes, hands, and feet, improving operational accuracy.

🧠 ABot-ER and ABot-AgentOS work together to form the top architecture, deepening decision-making on spatial and task relationships and unifying adaptation to heterogeneous robots.

8. Alibaba's Zhenwu M890 Super Node Successfully Compatible with Qwen3.8, Now Offers Inference Services on the BaiLian Platform

Alibaba's Zhenwu M890 super node successfully compatible with Qwen3.8, marking a significant breakthrough in domestic software and hardware collaboration in large model computing power scheduling, laying the foundation for the commercial deployment of next-generation large-scale AI applications.

【AiBase Summary:】

🚀 Alibaba's Zhenwu M890 super node successfully compatible with Qwen3.8, achieving a significant breakthrough in domestic software and hardware collaboration in large model computing power scheduling.

🧠 Qwen3.8 has a parameter scale of 2.4 trillion, and traditional AI clusters face performance bottlenecks, requiring distributed computing support.

🌐 Alibaba Cloud's BaiLian platform has launched this service, allowing enterprises and developers to directly call efficient trillion-parameter inference capabilities.