IEEE Spectrum AI

Prem Natarajan left Amazon to become Capital One's Chief Scientist, applying deep AI research to solve real-world financial challenges at scale, from fraud detection to agentic customer service.

This article explores how AI's rapid progress in mathematics challenges traditional research, prompting mathematicians to reassess their roles. It analyzes three potential futures: AI as a tool, a collaborator, or an autonomous oracle.

Princeton researchers are using reinforcement learning and generative AI to design radio-frequency integrated circuits (RFICs) from scratch, producing chips that outperform human designs in record time. The AI generates unconventional layouts that push performance limits, but the field needs open datasets to advance further.

Artificial intelligence, established as a distinct field at the 1956 Dartmouth Summer Research Project, has evolved over 70 years from early neural networks and expert systems to modern deep learning, large language models, and generative AI. The article reviews AI's history, strengths, risks, and IEEE's contributions to its progress and responsible use.

Large language models have moved from research labs into engineers' daily workflows. To help technical professionals stay ahead, IEEE offers a five-course online program that covers the engineering behind generative AI, from transformer architectures to deployment.

By mimicking how the brain operates, neuromorphic computing can use dramatically less energy than conventional electronic AI chips. However, even the most sophisticated neuromorphic devices today are still quite simple, using only a small fraction of the number of connections found in human neurons. Now a new study suggests that using sound waves, neuromorphic devices can better mimic biological neurons and operate faster and with greater energy efficiency than their electronic counterparts.

Generative AI complicates the definition of 'use' for musical works. Companies like Sureel and SoundVerse are developing attribution systems to track training data usage and ensure musicians are compensated when their work trains AI models. This raises technical, ethical, and policy challenges.

Researchers trained collaborative robots to read human emotions using vision language models, which outperformed traditional AI by incorporating context. However, while adaptive apologies were preferred, they could not repair trust lost due to task failure.

Isomorphic Labs, a Google DeepMind spinout, is using its novel AI system IsoDDE to discover hidden pockets on proteins for drug binding, going beyond AlphaFold. The system successfully predicted a cryptic pocket on cereblon, validating its ability to find novel drug targets.

Researchers at the University of Twente have shown that by adjusting GPU clock frequencies at the per-kernel level, they can save up to 14% of the energy used in LLM training with minimal impact on speed.

Tracking glacier loss is critical for climate research, but manual analysis is slow. A new AI approach reduces mapping error from over 1 km to under 70 m using minimal extra data: one hand-labeled image, summer reference images, and an underlying rock map. Applied to all 145 glaciers in Svalbard, it generated monthly calving front positions from 2015 to 2024. The method could help automate global glacier monitoring.

At Computex 2026, Nvidia announced RTX Spark, a version of its Blackwell GB10 superchip for Windows PCs. Microsoft and multiple OEMs unveiled devices. RTX Spark combines 20 Arm CPU cores, 6,144 GPU cores, and an NPU, targeting AI, gaming, and professional work. Nvidia's software dominance and industry clout may help establish Windows on Arm, but the challenge remains to compete with x86.

Quantum computers promise to solve problems beyond supercomputers, but their operation relies heavily on classical computing. As qubit counts rise, innovations in calibration and error-correction infrastructure are critical. Companies like NVIDIA, Q-CTRL, IBM, Riverlane, and Google are developing classical hardware and software to support quantum systems.

As AI systems become more capable, we pour resources into measuring their technical performance but largely ignore their psychosocial impacts on humans. Imran Khan of the Center for Humane Technology argues this is a critical oversight, with harms already appearing. He calls for long-term studies, data transparency, and regulatory liability to ensure AI helps humans flourish.

AI hardware startup Majestic Labs is developing a new AI server, Prometheus, with up to 128 terabytes of memory, over 60 times more than Nvidia's DGX B300. It uses a DRAM-centric architecture with a proprietary copper-cable memory interface and custom memory aggregation chips, delivering up to 25.6 TB/s bandwidth. The server features 12 Ignite AI processors combining ARM and RISC-V cores, and supports PyTorch, vLLM, and Triton frameworks without code modifications. Expected to ship in 2027, it claims to reduce capital expenditure and power consumption by 10 to 50 times.

An experienced ASIC designer who moved from academia to industry shares key insights on the differences between chip design in academic and industrial settings. The article covers divergent goals, risk tolerance, verification standards, and time horizons, and emphasizes the role of silicon IP. With the ASIC market growing rapidly, understanding these differences is crucial for aspiring chip designers from academia.

South Africa holds 88% of global platinum-group metals, hosts Africa's largest data center market, and sits at the center of a US-China AI infrastructure contest. Yet its draft AI policy, withdrawn after hallucinated references, fails to leverage these advantages for favorable terms. The article examines South Africa's structural leverage, three possible AI infrastructure futures (Chinese, US, local open-weight), and the need for binding governance provisions.

This webinar presents a workflow offering end-to-end solutions for designing, training, validating and verifying, compressing, and deploying AI-based virtual sensor models to embedded processors within a single environment.

Bees and other pollinating insects play vital roles in food webs and crop pollination, yet monitoring them has proven difficult. That’s why researchers have developed a radar system that could lead to a cost-effective, non-invasive way to track pollinators.

Researchers at the University of Waikato have developed a text-to-speech model for a dialect of te reo Māori, emphasizing data sovereignty and community ownership. Using open-source tools and minimal data, the model achieved a 6.78% word error rate and aims to provide a replicable blueprint for other minority language communities.

The open-source movement is bringing AI breakthroughs to robotics, lowering barriers to entry. From the ROS framework to models from Nvidia, Hugging Face, and Alibaba, robots' ability to reason, decide, and act is becoming accessible to more people. However, tensions between commercial incentives and academic ideals present new challenges.

Wetour Robotics argues that the next frontier in Physical AI is not smarter robots but smarter interfaces that treat the human body as a first-class computing node. Their Spatial Intent Fusion platform, Orchestra, combines spatial, visual, and gestural inputs to enable low-latency, hands-free control of connected devices. The system uses edge AI and sEMG for pre-motion intent sensing, addressing real-world scenarios where traditional interfaces fail. The article discusses the architecture, trade-offs, and implications for the field.

This presentation highlights recent efforts at the Johns Hopkins Applied Physics Laboratory to advance agentic AI for collaborative robotic teams. It begins by framing the core challenges of enabling autonomy, coordination, and adaptability across heterogeneous systems, then introduces a scalable architecture designed to support agentic behaviors in multi-robot environments. The talk concludes with key challenges encountered and practical lessons learned from ongoing research and development.

Melbourne is leveraging sovereign AI compute, hyperscale data centers, and international conferences to create a research flywheel. MAVERIC supercomputer, investments by CDC and NEXTDC, and upcoming conferences position Melbourne as a global hub for AI research.

Research reveals that AI voice systems can be hijacked via imperceptible audio signals, with success rates up to 96%. The AudioHijack attack targets large audio-language models (LALMs) and can trigger malicious actions including data theft and unauthorized tool use.

Electronic rings wirelessly connected to an AI system are capable of translating multiple sign languages into text, a new study finds. The system uses seven rings with accelerometers to detect hand motion, achieving 88.3% and 88.5% accuracy for 100 common ASL and International Sign Language words, and can translate continuous sentences. Future work includes incorporating facial expressions and body posture, and migrating processing to smartphones.

Researchers at the University of Texas at Austin have developed a graphene tattoo that sticks directly onto leaves to monitor hydration in real time. The sensor acts as an artificial synapse, potentially enabling plant-based neural networks for forest fire and drought monitoring.

A new study in Science shows OpenAI's LLM outperforms physicians on clinical reasoning tasks, but sparks debate over reliability, evaluation standards, and the path to responsible use in healthcare.

General-purpose LLMs are increasingly capable of transcribing historical handwriting, outperforming specialized tools like Transkribus in accuracy, speed, and cost, opening up previously inaccessible archival collections.

As AI workloads reach gigascale levels, data centers face a physical bottleneck in power chain resilience. GPU clusters create high-frequency pulse loads that legacy systems cannot handle. Ampace and Eaton collaborate to transform energy storage from passive backup to active stabilization using semi-solid batteries and intelligent UPS systems.