Neural Chip Decodes Interactions Between Brain Regions

A low-power implantable neural system-on-chip developed by researchers at the École Polytechnique Fédérale de Lausanne (EPFL) in Switzerland monitors communication between brain regions to identify cognitive and emotional states in real time. The LogSync technology uses machine learning to analyze brain-wave strength and synchronization, generating triggers for precisely timed electrical stimulation. The chip consumes less than 200 microwatts, enabling complex neural processing within the limited power budget of an implant.

More info here: EPFL News (Switzerland); Celia Luterbacher

U.S. Designates Helium a Critical Material, Citing Importance for Semiconductors

The U.S. Department of Energy has designated helium as a critical material because it is essential for semiconductor manufacturing, advanced computing, and other vital needs while global supplies face disruption. The designation increases access to federal financing for producers and strengthens government monitoring. The U.S. produced 43% of global helium supply in 2025, but domestic output is constrained by extraction plants operating at capacity and lengthy permitting processes.

More info here: Associated Press; Jennifer McDermott; Alexa St. John

Professor Creates Python-like Language for AI Agents

University of Michigan professor Jason Mars developed Jac, a Python-like programming language, and Jaseci, a runtime designed to make AI agents more efficient. Jac integrates Python and TypeScript ecosystems, supports libraries from PyPI and npm, and compiles to Python bytecode, JavaScript, and native machine code. Its graph-based data structures allow agents to navigate relationships and process information directly.

More info here: The Register (U.K.); Joab Jackson

Stealthy Low-Cost Tech Detects Drones

A low-cost drone detection system developed by Florida International University researchers identifies the magnetic signatures produced by spinning electric motors. Using a specialized antenna and machine learning, the technology identified outdoor drones by make and model with up to 98.52% accuracy at distances of up to 15 meters. Unlike radar, the system passively detects signals without transmitting emissions, reducing the risk of detection by adversaries.

More info here: FIU News; David Drucker

Nobel Laureate, Team Use AI to Solve Decade-Old Math Puzzle

Nobel Prize-winning physicist Giorgio Parisi and colleague Francesco Zamponi used generative AI to help solve a mathematical puzzle in physics that had remained unresolved for more than a decade. The problem involved «jamming,» the transition in which a disordered system, such as tightly packed billiard balls, becomes rigid. Researchers had observed that two key parameters always summed to one at the jamming point but could not prove why. Parisi and Zamponi used Anthropic’s Claude AI to help uncover a path toward a proof.

More info here: LiveScience; Benjamin Skuse

Ukrainian Cruise Missile Flies with Hobby Hardware

A low-cost cruise missile developed by Ukraine uses the open-source ArduPilot flight controller, which can run on low-cost hardware and is commonly used by drone hobbyists. By relying on inexpensive commercial hardware and community-developed software, Ukraine has reduced costs while maintaining effective long-range capabilities. Russia is similarly using low-cost commercial components, including Raspberry Pi processors, in newer attack drones with AI-assisted targeting.

More info here: New Scientist; David Hambling

China’s Moonshot unveils world’s largest open AI model

Chinese AI startup Moonshot AI has unveiled a 2.8-trillion-parameter AI model that it said is the world’s largest open-weight AI system. The Kimi K3 model is designed for advanced reasoning, long-horizon coding, and knowledge work, and features a 1-million-token context window. Moonshot said Kimi K3 approaches the performance of leading U.S. frontier models; third-party evaluations ranked it highly on Web interface-building and complex multistep tasks.

More info here: Reuters; Laurie Chen

IBM Claims World’s First Sub-1 Nanometer Chip Technology

IBM unveiled a chip manufacturing technology that it says could extend semiconductor miniaturization, countering concerns that advances in shrinking transistors are nearing their limits. The company’s new «nanostack» architecture vertically bonds two layers of nanosheet transistors, nearly doubling transistor density on a fingernail-sized chip compared with its 2021 design. IBM says the technology could deliver 50% higher performance or 70% better energy efficiency.

More info here: Ars Technica; Jeremy Hsu

DOE Plans to Create Scientifically-Relevant Quantum Computer

The U.S. Department of Energy (DOE) launched Quantum Genesis, an initiative aimed at developing the world’s first scientifically relevant, fault-tolerant quantum computer. The program seeks to build a system with 150–250 logical, error-corrected qubits by 2028, enabling advanced scientific research and complex calculations beyond current computing capabilities. The initiative will bring together national laboratories, universities, and private-sector partners while establishing a new user facility that provides researchers access to quantum computing resources alongside AI and high-performance computing tools.

More info here: Nextgov/FCW; Alexandra Kelley

Papyrus Scroll Burnt During Vesuvius Eruption Deciphered by AI

Researchers used AI to virtually unwrap and read a charred Herculaneum scroll that was buried during the eruption of Mount Vesuvius in AD79. Without physically opening the fragile papyrus, scientists revealed 20 columns of previously hidden text discussing Stoic philosophy, ethics, reason, and human behavior, likely dating to the second or third century BC. The breakthrough was achieved through the global Vesuvius Challenge contest, and combines AI and high-resolution X-ray imaging to decipher carbonized scrolls.

More info here: The Guardian (U.K.); Ian Sample