News Summaries

How AI Agents Are Learning to Reuse Skills Like Humans – The Skill‑Pro Breakthrough

Researchers at the Institute of Automation unveiled a new system called Skill‑Pro that lets large‑language‑model (LLM) agents remember and reuse useful tricks instead of starting from scratch each time they face a problem. Imagine a robot that, after solving a puzzle once, can pull out the exact steps it used and apply them to a similar challenge later. Skill‑Pro does this by turning fragments of past interactions into tidy, executable “skills” that include clear start conditions, step‑by‑step instructions, and stop signals. The clever part is that the system doesn’t need to retrain the whole model; it uses a non‑parametric version of the popular PPO algorithm to generate and test candidate skills, keeping only the most reliable ones. A built‑in “gate” checks each skill for safety, while a scoring system trims the library so it stays compact and high‑quality. Tests showed that agents using Skill‑Pro could solve tasks faster, use far less memory, and handle new problems they hadn’t seen before. Visualizations of skill evolution revealed how the agents gradually build a toolbox of knowledge, paving the way for AI that can learn, refine, and reuse its own expertise over the long term.

SpaceX Rockets 29 New Starlink Satellites Into Sky From Florida

On the morning of May 29, 2026, SpaceX lifted off from Cape Canaveral’s Launch Complex 40 at 8:57 a.m. EDT, sending a Falcon 9 rocket soaring into the Atlantic. The mission, dubbed Starlink‑29, carried 29 brand‑new broadband‑service satellites destined for low‑Earth orbit. Once deployed, the satellites will join an ever‑growing constellation that aims to deliver high‑speed internet to remote and underserved regions around the globe. The Falcon 9’s first stage performed a controlled landing back at the nearby drone ship, marking another successful reuse of the rocket’s hardware. Inside the payload fairing, each Starlink satellite is equipped with advanced phased‑array antennas and high‑throughput transponders, allowing the network to handle more users and provide faster data rates than earlier batches. SpaceX’s rapid launch cadence reflects its ambition to reach a target of several thousand operational satellites within the next few years, a goal that could reshape global connectivity and support emerging technologies such as autonomous vehicles and remote‑sensing applications. The Florida launch underscores the state’s continuing role as a hub for commercial space activity, and it adds another milestone to SpaceX’s record of reliable, high‑frequency missions. Fans and industry watchers celebrated the successful liftoff on social media, noting the smooth ascent, the precise deployment of the payload, and the continued progress toward a truly global internet service.

Chinese Researchers Harness AI to Supercharge Carbon Dioxide Power Technology

At the China Nuclear Power Research and Design Institute in Chengdu, a team of young scientists has achieved a major breakthrough in supercritical carbon‑dioxide (s‑CO₂) power generation, a technology that could make electricity plants smaller, cleaner and more efficient. The “Supercarbon One” project, driven by graduate students and seasoned mentors, spent months on round‑the‑clock testing, often sleeping on the lab bench and missing family gatherings to fine‑tune experimental setups and data models. Now the group is turning to artificial intelligence to push the technology further. They are building an AI‑driven simulation platform that lets engineers talk to the software in natural language, instantly see real‑time plant conditions, predict faults before they happen and automatically adjust operating parameters for optimal performance. The system combines large‑scale language models with specialized industrial tools, creating a digital twin of the s‑CO₂ cycle. The work supports China’s 15th Five‑Year Plan, which calls for breaking energy‑sector bottlenecks and strengthening domestic control of critical supply chains. Researchers aim to scale the technology to higher power levels, wider temperature ranges and tougher operating environments, accelerating its commercial rollout. Their story highlights the blend of perseverance, patriotism and cutting‑edge innovation shaping the next generation of clean energy.

Stacked‑Silicon Chip Could Keep Computers Getting Faster for Years

Researchers at the University of Illinois have found a way to keep the pace of faster, smaller computers moving forward, even as traditional chip‑shrinking methods start to stall. Their new technique stacks ultra‑thin layers of silicon circuits on top of each other, creating a true three‑dimensional chip. By using a low‑temperature manufacturing process, the team overcame a long‑standing hurdle that made building multi‑layer chips difficult and costly. The result is a chip that packs far more computing power into the same footprint, runs cooler, and uses less electricity. This breakthrough could extend the trend famously known as Moore’s Law— the observation that the number of transistors on a chip doubles roughly every two years— for many more years. In practical terms, the technology promises faster smartphones, more responsive laptops, and smarter devices without the need for larger batteries or bigger hardware. While still in the research stage, the approach shows promise for commercial production, potentially reshaping the semiconductor industry and keeping the rapid pace of digital innovation alive.

China’s Robot Boom: 90% Market Share, but Reliance on U.S. Tech Raises Concerns

Generative AI and large language models are sparking a wave of AI‑powered robotics in China. Roughly 150 Chinese firms are building humanoid robots, each focusing on a different piece of the puzzle—some perfect motor skills, others train robots with AI models that can see, talk and act. While the sector is buzzing with innovation, it’s still in its infancy and heavily leans on U.S. breakthroughs in vision‑language‑action (VLA) technology. American companies score an "A" for computing power, data handling and robotic hands, leaving Chinese makers playing catch‑up. Chinese startups like Zhiyuan Robotics, Galaxy Universal and Zhifang Technology have launched their own VLA‑inspired models, but these are narrow‑focused and far from fully autonomous. Parallel research into "world models"—AI that builds a 3‑D understanding of its surroundings—is underway at labs led by AI pioneers such as Fei‑Fei Li and Yann LeCun, with big players like Google DeepMind, Meta and NVIDIA also in the race. On the industrial front, China now dominates the global robot market, installing more machines than the rest of the world combined. In 2024, domestic firms satisfied over 57 % of local demand, edging out foreign rivals. Yet most of these robots are task‑specific arms, not the humanoid companions that capture headlines. Europe watches closely, aware that its strong manufacturing base must contend with China’s massive supply chain and its lingering dependence on U.S. AI hardware. The report warns that while China’s robot sector is booming, its reliance on U.S. technology—especially NVIDIA’s AI stack—remains a strategic vulnerability.

Ancient Protein Map Uncovers Hundreds of Mystery Genes Behind Rare Diseases

A team of scientists has pieced together a 1.5‑billion‑year‑old protein interaction network—essentially a molecular map from the dawn of complex life—to hunt for hidden genes that may cause human disease. By comparing this ancient blueprint with modern genomes, they identified hundreds of previously unknown genes that appear linked to a variety of health conditions. To test their findings, the researchers turned to frog and mouse models, where they confirmed that three of the newly flagged genes are indeed tied to rare disorders: osteopetrosis (a bone‑hardening disease), end‑stage kidney disease, and short‑rib thoracic dysplasia, a severe skeletal malformation. These early successes suggest the ancient network can spotlight disease‑relevant genes that modern studies have missed. The team plans to expand their work, using additional animal experiments to verify more candidate genes and explore how these ancient connections influence today’s biology. If successful, this approach could open a new frontier in genetics, offering fresh targets for diagnostics and therapies and shedding light on the deep evolutionary roots of human disease.

Scientists Unveil a Gold Mine of Black Hole Collisions, Sharpening the Hunt for the Universe’s Expansion Rate

Astronomers have just added a spectacular new find to their growing catalog of gravitational‑wave events: a pair of black holes, each about 26 and 30 times the mass of our Sun, slammed together more than 3 billion light‑years away. The collision produced a ripple in space‑time that was pinpointed with unprecedented accuracy, giving researchers a clearer picture of where the event happened. This discovery is part of the latest GWTC‑5.0 catalog, which now holds dozens of such cosmic crashes. With each new detection, scientists gain a fresh way to measure how fast the universe is expanding—a question that has puzzled cosmologists for decades. By comparing the distance inferred from the gravitational wave signal with the redshift of the host galaxy, they can calculate the Hubble constant without relying on traditional methods like supernovae or the cosmic microwave background. "The updated catalog gives us a much larger collection of gravitational‑wave signals to help answer one of the biggest questions in cosmology: how fast is the universe expanding?" said Alex Papadopoulos, a postgraduate researcher at the Institute for Gravitational Research. As the treasure trove of black‑hole mergers grows, so does our ability to weigh the cosmos with ever‑greater precision.