Digital News

　　Recently, Shanghai once again became the global hotspot for the AI community as the 2025 World Artificial Intelligence Conference (WAIC 2025) kicked off at the Shanghai Expo Center under the theme “Global Solidarity in the AI Era.” Over 800 companies gathered in a venue spanning more than 70,000 square meters to showcase over 3,000 cutting-edge exhibits to the world. These included more than 40 large models, over 50 AI terminal products, more than 60 intelligent robots, and over 100 new products making their global or China debut. In terms of exhibition size, number of exhibits, and the scale of knowledge co-creation, this event set unprecedented records.

　　Currently, a global surge in intelligent technology is reshaping the landscape, with the AI field experiencing a wave of collective breakthroughs. Innovations in large language models, multimodal large language models, and embodied intelligence are evolving at a breathtaking pace, rapidly propelling AI toward greater efficiency and more advanced capabilities. AI today is no longer just a combination of computing power, algorithms, nor is it a distant or abstract concept. Instead, it is quietly weaving itself into the fabric of countless industries, creating meaningful impact in ways that are subtle yet profound, and becoming more tangible and accessible in everyday life.

　　At the H1 Core Technologies Pavilion, StepFun unveiled its next-generation foundational model, Step 3. In partnership with Qianli Technology and Geely Auto, it also launched a preview version of Agent OS—a next-generation intelligent cockpit system designed for AI agents. The project earned the WAIC “Pavilion Highlight” Award for excellence in applied innovation. Based on StepFun’s industry-leading multimodal large language model and end-to-end voice large model, this intelligent cockpit can recognize passengers’ emotions in real time and adapt the in-car environment accordingly. For example, if the system detects that the user is on a phone call, it will automatically turn down the music volume. Using memory-based navigation, the vehicle can suggest routes to the user’s frequently visited restaurants, transforming the cockpit from a functional space into a personalized “third living space.”

　　According to reports, over the past two years, StepFun’s self-developed Step series large models have been incorporated into flagship products of leading companies across the automotive, embodied intelligence, and IoT sectors. Moreover, more than half of the top domestic smartphone brands have partnered with StepFun to integrate AI agents. Meanwhile, StepFun is actively expanding into vertical industries, partnering closely with leading companies in finance, content creation, retail, and more to jointly develop scenario-based, consumer-centric experiences. The commercial achievements have driven robust revenue growth. Building on rapid growth in the first half of the year, StepFun aims to achieve an annual revenue target of 1 billion yuan.

　　At the SenseTime booth, the AI companion robot “Fuzozo” from Robopoet, equipped with the Multimodal Emotion Model (MEM) and a biomimetic memory system, drew considerable attention from visitors. “This product is built around AI large model technology, providing users with emotional support and companionship,” a staff member explained to the reporter. It enhances user engagement and fosters a sense of belonging through a variety of nurturing systems and social interactions. “For instance, when I gently touch its head, it responds with biomimetic vibrations as feedback.”

　　Additionally, at the WAIC 2025 Large Model Forum, SenseTime launched its “Wu Neng” embodied intelligence platform. During the event, SenseTime’s Chairman and CEO, Xu Li, showcased a humanoid robot equipped with the embodied world engine. The robot can autonomously plan routes based on instructions such as “find something on the kitchen shelf” and engage with the real world by generating multi-view videos. During the demonstration, the robot even grasped the cultural metaphor behind the film The Litchi Road, smoothly advanced presentation slides, and answered questions from the audience, showcasing an impressive ability to follow instructions. Xu Li shared that SenseTime envisions the “Wu Neng” embodied intelligence platform as a tool to help companies in the field realize the dream of interacting with the physical world.

　　What the world looks like has always been a proposition in human exploration. If you’ve ever wondered how to capture a nearly 4-meter-long artwork like the Complete Map of All Nations on Earth (Kunyu Wanguo Quantu) on your phone. CamScanner, a product of INTSIG Information, has launched “infinite scanning” technology, featuring “large-format scanning,” which may be helpful. Backed by AI scanning technology, the reporter noticed that users only need to swipe up, down, left, and right while recording a video to capture the entire large map in high definition and with just one tap, seamlessly save it to their phone. Using AI-powered image recognition, CamScanner can intelligently identify elements within the Complete Map of All Nations on Earth, such as mountains, territories, and mythical animals. By tapping on a designated area, users can explore how people viewed the world more than 400 years ago, delivering a digital experience that truly “lets artifacts speak and gives history a voice.” A staff member at the booth explained that this technology enables users to effortlessly preserve engineering drawings, large hand-drawn sketches, and more. It also allows for high-definition captures of wall paintings, murals, and other images while traveling or visiting exhibitions, making it easy to share professional artwork with friends and helping art reach a wider audience without boundaries.

　　At last year’s WAIC, 18 humanoid robots lined up to welcome attendees, sparking heated discussion. This year, the WAIC continued to dazzle with an impressive lineup of humanoid robots. Unitree Robotics’ G1 combat robot, equipped with 29 joints, presented an “AI Bruce Lee” performance, attracting the audience with its smooth and powerful spinning kicks in the ring. KEENON Robotics designed their exhibition area into a bar, a restaurant, and a theater, where their bipedal service robots, XMAN-F1, performed role-based services like popcorn making, cold drink mixing, tour guiding, and food delivery, forming a “robot service matrix.” The full-sized bipedal humanoid robot “Xingdong L7,” presented by Robot Era, wowed the crowd with a Breaking dance routine showcasing 55 degrees of freedom. It then smoothly transitioned into logistics sorting mode, precisely scanning package barcodes and logging them into the system—effortlessly shifting from performance to productivity.

　　The empowering effects of AI are becoming increasingly apparent in key sectors such as industry and finance. SmartEase, NetEase’s intelligent construction machinery brand, has introduced “Lingjue,” the world’s first embodied intelligence model specifically developed for construction machinery. Through the intelligent cockpit, visitors can remotely experience the autonomous loading operation of unmanned excavators working thousands of miles away at a mining site in Inner Mongolia, bringing AI technology from the office straight to the rugged, dust-filled mines. According to reports, “Lingjue,” built on a fully domestically developed training framework called “Jixie Zhixin,” achieved a single-machine loading efficiency of 80% compared to manual operations, and about 70% of the working time required no human intervention, demonstrating remarkable stability and environmental adaptability during field tests at an open-pit coal mine in Inner Mongolia.

　　In the realm of financial risk control, INTSIG Information’s AI-powered counterfeit detection technology acts as a “firewall” to safeguard information authenticity. At the booth’s interactive area, after scanning the reporter’s face, the system instantly generated a real-time “digital mask” and, within milliseconds, identified the proportion of forged pixels by analyzing subtle visual discrepancies invisible to the human eye. For AIGC-generated images, the multimodal model analyzes the rationality of lighting, shadows, and the perspective relation. Even when affected by compression or cropping, it still maintains an identification accuracy of over 90%.

　　AI is evolving at an accelerated pace, entering a phase of explosive growth. As it moves from the lab into a wide range of industries and daily life, it is reshaping ecosystems and emerging as a new driver of economic growth. Perhaps the future has arrived.