A team of researchers has unveiled a surprisingly inexpensive method for catching "hallucinations"—the false or fabricated statements that large language models sometimes generate. While these AI systems can produce impressively fluent text, they occasionally invent facts or misinterpret prompts, which can erode user trust and even cause real‑world problems when the output is taken at face value. The new approach uses a lightweight statistical filter that monitors the model’s internal confidence scores and cross‑checks generated claims against a compact, curated knowledge base. By flagging inconsistencies in real time, the system can alert developers or end‑users before erroneous information spreads. Because the technique relies on simple probability thresholds and a modest reference dataset, it can be deployed on standard hardware without the need for costly compute resources. Early tests show that the method catches up to 85 % of hallucinated statements while only slowing down the model’s response time by a few milliseconds. This cost‑effective solution promises to make AI applications—from chatbots to automated report writers—more reliable and safer for everyday users, marking a significant step forward in responsible AI deployment.
Read moreIn a groundbreaking experiment, researchers at Lawrence Livermore National Laboratory (LLNL) melted a tiny diamond under pressures millions of times greater than Earth’s atmosphere. By doing so, they recreated the extreme conditions found deep inside ice‑giant planets such as Neptune and Uranus. The molten diamond acted as a transparent window, allowing scientists to watch how water, ammonia, and other “ice” compounds behave when squeezed to the core of these distant worlds. The findings show that under such crushing forces, these ices turn into exotic, metallic‑like fluids that conduct electricity and generate magnetic fields—insights that could finally explain the puzzling magnetic anomalies observed on the real planets. Beyond planetary science, the experiment has a surprising link to the quest for clean energy. LLNL also houses the National Ignition Facility, where teams are trying to achieve net‑positive nuclear fusion using inertial confinement. Understanding how matter behaves at ultra‑high pressure helps refine the models that guide fusion capsule designs, bringing us a step closer to a sustainable fusion power source. While we won’t see a diamond‑melting lab on TV anytime soon, the research bridges the mysteries of the outer solar system with the practical goal of harnessing star‑like energy here on Earth.
Read moreA team of physicists has shown that a single, atom‑thin chip can detect magnetic fields from multiple angles at once. By stacking ultra‑precise layers of two‑dimensional quantum materials, they created a heterostructure that behaves like a tiny, customizable sensor. In their experiments the device displayed the familiar Hall effect—a voltage that appears when a magnetic field pushes electric charges sideways—but it also produced an unexpected signal that follows the magnetization lying in the plane of the chip. In plain terms, the chip can read magnetic fields both perpendicular to its surface and parallel to it, something no conventional Hall sensor can do. The discovery, reported in *Nature Materials*, highlights the power of engineering materials atom by atom to unlock new electronic and magnetic functions on demand. Lead researcher Katoch emphasized that this work “demonstrates the power of building atomically precise heterostructures of emergent two‑dimensional quantum materials to obtain on‑demand electronic and magnetic properties.” Potential applications range from ultra‑compact navigation systems and medical imaging tools to next‑generation data storage that relies on precise magnetic control. As the technology matures, we could see a new class of multi‑axis magnetic sensors that are thinner than a human hair and far more versatile than today’s devices.
Read moreA team of researchers has demonstrated a way to beam data through the air with laser light while keeping it almost completely intact. By wrapping the light signal in tiny magnetic whirlpools called “skyrmions,” they were able to block the usual distortion that plagues free‑space communications, achieving a 98 % fidelity rate across distances of several hundred metres. Unlike traditional fiber‑optic cables, this method needs no physical wiring, which could dramatically cut the cost and complexity of high‑speed links—especially in places where laying fiber is impractical, such as remote villages, disaster zones, or outer‑space missions. The breakthrough also promises to boost emerging technologies like satellite broadband, quantum computing networks, and even future space‑to‑Earth data streams. The scientists say the result shows that light’s natural, distortion‑resistant properties can be harnessed for fast, reliable communication without constant monitoring or correction. If the technique scales up, we could see “laser internet” services similar to Starlink delivering gigabit speeds to underserved areas, all without the need for costly cable infrastructure. The work opens a new path toward cheaper, more flexible global connectivity.
Read moreOn August 24, the State-owned Assets Supervision and Administration Commission (SASAC) held a high‑level meeting to accelerate the rollout of 6G across China’s central enterprises. The gathering reviewed progress on the Communist Party’s 6G directives, set out the next phase of work, and stressed that mastering 6G is essential for meeting national strategic goals. Key take‑aways: 1. **Technology breakthroughs** – firms must close gaps in basic theory, develop core components, and secure a stable supply chain for 6G hardware. 2. **Industrial foundation** – expand innovation platforms, nurture market players, and build talent pools to create a robust 6G ecosystem. 3. **Global cooperation** – deepen partnerships abroad to boost China’s influence in international standards and technology sharing. 4. **Real‑world applications** – co‑create use‑case scenarios that embed 6G into sectors such as manufacturing, transportation, aerospace, and low‑altitude services, aiming for a closed‑loop commercial model. SASAC data show central enterprises will invest 2.5 trillion yuan in strategic emerging industries by 2025, with emerging‑industry revenue expected to rise over 10 percentage points from 2022 levels. Shanghai’s 15th Five‑Year Plan also pledges to lead 6G deployment, building a “air‑space‑land‑sea” communication network and integrating ultra‑large‑scale intelligent computing clusters. The push signals a coordinated effort to turn 6G from a research concept into a driver of economic growth and industrial modernization.
Read moreWuxi’s High‑Tech Zone is fast becoming China’s aerospace hotspot. By gathering more than 30 specialised firms—including launch‑vehicle pioneer Deep Blue Aerospace, satellite‑maker Guoyu Xingkong, and heavy‑load drone builder Muyutian Aviation—the district has woven a full‑stack industry chain that now exceeds 10 billion yuan in value. Highlights this year include Deep Blue’s successful integration test of its reusable “Nebula‑1” rocket and the rollout of Muyutian’s 8‑ton AT8000 “flying heavy truck.” Guoyu Xingkong’s new production line can churn out 200‑500 kg satellites at a rate of 150 per year, cutting assembly time to as little as 25 days. The zone also backs an 18‑billion‑yuan commercial‑aerospace fund and offers fast‑track financing, cutting a rocket‑assembly plant’s set‑up time to six months. Real‑world pilots are already showing the impact: drones patrol city roads, a “instant‑response + drone‑delivery” model boosts logistics speed by 50 %, and satellite remote‑sensing powers precise environmental monitoring. With 28 municipal‑level R&D labs, partnerships with top universities, and a focus on chips, operating systems and ultra‑precision components, Wuxi is turning lofty space dreams into everyday services while laying the groundwork for the next generation of aerial commuting and on‑orbit manufacturing.
Read moreA new AI‑driven platform called the Nenghuanbao NiOS Smart Energy Management System is reshaping how solar power plants are run. Using a digital‑twin model that mirrors each plant in real time, the system predicts how much electricity will be generated with an error margin of less than 5% and spots equipment problems with over 95% accuracy. It pulls in satellite weather data, analyzes sunlight patterns, and alerts operators before a fault can cause a loss. The platform links drones, ground‑based robots and infrared scanners into an “air‑ground‑space” inspection network. Drones equipped with thermal cameras locate hot spots or dust build‑up on panels, while autonomous robots clean and service the arrays, reducing the need for risky high‑altitude work. Operators interact with the system through voice commands and natural‑language reports, letting a small team monitor dozens of plants from a single dashboard. Automation handles data collection, fault alerts and work‑order dispatch every minute, slashing labor costs and speeding up repairs. By forecasting inverter wear and adjusting energy‑storage strategies ahead of storms, the system keeps power flowing even in extreme weather. Deployed at more than 100 projects nationwide, NiOS has already boosted generation efficiency, lowered carbon emissions and demonstrated how AI can make renewable energy cleaner, cheaper and more reliable.
Read moreA team of scientists from the Chinese Academy of Sciences and Singapore’s Nanyang Technological University has created a breakthrough sensor that lets robots perceive the world the way humans do – by seeing and feeling at the same time. Traditional robots use separate cameras for vision and pressure pads for touch, which often leads to mismatched data and slower, less accurate decisions. Inspired by the sea‑urchin’s tube feet, where light‑detecting and pressure‑detecting cells live side‑by‑side, the researchers built a single, flexible material that combines an optical layer and a pressure layer in one compact unit. The sensor captures light and force signals simultaneously, eliminating the need for complex post‑processing and reducing signal interference. In laboratory tests, robots equipped with the new device identified objects with a 96 % success rate, even in dynamic or cluttered environments. Published in *Nature Sensors*, the work promises to accelerate advances in embodied intelligence, human‑machine interaction, and precision manufacturing, offering a more natural, “homogeneous” way for machines to understand their surroundings.
Read moreArtificial‑intelligence agents are moving from research labs into everyday business. Unlike a single‑turn chatbot, an AI agent can see, think, plan, remember, use tools and act on its own to reach a goal. In China, a state‑owned bank rolled out a credit‑risk agent that now drafts loan‑review reports 17 times faster and cuts human checks by two‑thirds. A top hospital’s medical‑guideline agent lifted treatment‑plan compliance from 73% to 96% and slashed medication errors by 89%. A manufacturing group’s supply‑chain agent warns of delivery risks in minutes across 237 suppliers and even starts negotiations automatically. Over the next three years, agents will become tiny (under 50 MB) enough to run on IoT devices, highly specialized for sectors such as health, finance or government, and equipped with built‑in explainability, fact‑checking and blockchain proof. Six main camps dominate the domestic market: internet giants (Baidu, Alibaba, Tencent, ByteDance) offering full‑stack platforms; cloud providers (Huawei, Alibaba Cloud, Tencent Cloud) delivering secure, private runtimes; traditional software vendors embedding agents into ERP and CRM; low‑code/RPA firms turning scripts into “cognitive RPA”; AI startups pushing cutting‑edge architectures; and vertical‑focused players pre‑loading industry knowledge graphs and compliance tools. Together, these platforms form a new digital nervous system that could decide how China competes in the global AI‑agent era.
Read more