This article takes a big‑picture look at the rapid rise of massive AI models and the edge‑computing tech that powers them. It starts by explaining AI’s evolution—from simple rule‑based systems (weak AI) to the dream of human‑like intelligence (strong AI) and beyond. Recent forecasts from Gartner show that 2024 marks the first year of “everyday generative AI,” when tools that can write, draw, and reason are moving out of labs and into homes, shops, and factories. The piece highlights DeepSeek, a home‑grown large‑model champion, and its breakthroughs: longer memory windows, faster reasoning, and lower response times that make complex tasks feel instant. DeepSeek also packs more useful knowledge per parameter thanks to careful data cleaning and advanced training tricks. Across industries, the article maps how AI is already at work—personalized product recommendations, streaming‑service suggestions, scientific research assistants, design tools for architecture and aerospace, facial‑recognition security, medical‑image analysis, energy‑saving systems, and self‑driving cars. Decision‑making AI powers rule engines and content moderation, while generative AI fuels a new wave of content creation, code writing, and multimodal (text‑image‑audio) understanding. Looking ahead to 2025, the focus shifts to AI governance platforms and autonomous agents that can make decisions, execute tasks, and collaborate with other systems without human oversight. In short, massive models and edge computing are turning AI from a niche research topic into a daily utility that reshapes how we work, create, and live.
Read moreChina’s biggest AI labs—DeepSeek, Zhipu and Moonshot—are locked in a fierce battle to build the next generation of AI infrastructure. Their latest model, Kimi K3, is already matching or beating some of the world’s most advanced systems, such as Anthropic’s Fable 5, signaling that Chinese models are moving from “catch‑up” to “keep‑pace.” Yet experts say true breakthroughs still lie ahead. The original dream of the 1956 Dartmouth Conference—to create machines that can learn from a single example, explain their reasoning, and independently generate scientific ideas—remains unfinished. Future competition will focus on continuous learning, multimodal abilities (combining text, images, sound) and AI‑driven discovery. Beyond raw power, AI is reshaping how we work. “Agents” are turning programming into a skill anyone can pick up, as long as they can define a problem, test solutions, and set clear success criteria. According to veteran technologist Hong Xiaowen, the real value of AI isn’t just what it can do, but how it forces us to rethink human intelligence: from building things to deciding what to build, why, and who’s responsible. He boils intelligent systems down to three ingredients—algorithms, data, and compute—and stresses that language is the glue that lets machines tap the world’s knowledge. In short, AI is moving from a philosophical curiosity to a new economic engine that could reshape society.
Read moreIn 2026 a team from Nanjing University, working with Suzhou’s National Lab and Huawei, turned a long‑standing research dream into a working 2‑D semiconductor chip. They built the world’s first molybdenum‑disulfide (MoS₂) multi‑bit processor, called “Mengqi‑1000”, packing 1,433 transistors onto a tiny 0.5 µm area. The chip runs on a standard RISC instruction set, reaches 43 kHz, and stores data directly on the 2‑D layer, eliminating the slow, power‑hungry off‑chip memory bottleneck. Its transistor density—over 9,300 per mm²—matches mature silicon technology, marking a major step toward mass production. At the same time, researchers at Fudan University created a 2‑D transistor with record‑low leakage, losing only a single electron every 9.15 seconds. Using this, they built a DRAM‑like memory that can hold data for more than 8,500 seconds without power, writes in nanoseconds, and offers 5‑bit precision. The ultra‑low leakage could slash refresh power for edge‑computing and high‑performance AI chips, especially when combined with 3‑D stacking. A breakthrough in p‑type 2‑D materials also arrived: the Chinese Academy of Sciences grew large‑area, high‑mobility MoSi₂N₄ wafers, solving a key hurdle for full 2‑D CMOS circuits. Parallel advances in ferroelectric thin films from Peking University delivered transistors that switch at just 0.8 V and survive over a trillion cycles, promising ultra‑efficient future chips. Together, these material, device, and equipment innovations are knitting a complete 2‑D semiconductor ecosystem that is moving rapidly from university labs to commercial fabs.
Read moreOn July 19, more than 100 leading researchers, university professors, and industry innovators gathered in Beijing for a special forum on "Embodied Intelligent Robots and Swarm Intelligent Agent Collaboration" held during the 28th China Association for Science and Technology Annual Conference. Organized by the Chinese Automation Society, the Chinese Society of Particle Technology, and the Shenyang Institute of Automation, the event spotlighted breakthrough ideas that could reshape how robots think, learn, and work together. Keynote speakers presented bold new concepts: Xu Kai introduced a hybrid model that blends physical‑world knowledge with data‑driven AI, while Lin Liang unveiled PhyAgentOS, an operating system that separates planning from execution for smarter robot behavior. Ji Rongrong showcased a multimodal large‑model architecture that speeds up robot perception, and Lin Zhouchen revealed a weight‑free tuning technique that sidesteps traditional bottlenecks. Other highlights included Xiong Rong’s vision of deep data fusion for general intelligence, Sun Fuchun’s self‑growing robots that reshape themselves to fit environments, and Chen Qijun’s end‑to‑end navigation system for mobile robots in unknown spaces. A round‑table discussion debated the value of human‑generated data versus autonomous learning, sparking fresh research directions. The forum not only fostered cross‑disciplinary dialogue but also set a roadmap for China’s future in robotics and artificial intelligence, promising smarter, more adaptable machines for everyday life.
Read more