Keumars Afifi-Sabet

Scientists have built a framework that gives generative AI systems like DALL·E 3 and Stable Diffusion a major boost by condensing them into smaller models — without compromising their quality.

New chip design uses photons rather than electrons to perform calculations, and scientists hope to integrate the technology into future graphics cards to train AI.

This 14-inch wireless charging device works at a rate of 100 kW, scientists claim, meaning it's up to 10 times as fast as some of the best commercially available alternatives.

Scientists trained an AI system to think before speaking with a technique called QuietSTaR. The inner monologue improved common sense reasoning and doubled math performance.

By letting different processing units — like GPUs, NPUs and hardware accelerators — work in parallel, rather than in sequence, systems can be up to twice as fast and consume 50% less energy.

Scientists built an app that let them control a robot using hand gestures while wearing the Apple Vision Pro VR headset.

Cerebras' Wafer Scale Engine 3 (WSE-3) chip contains four trillion transistors and will power the 8-exaFLOP Condor Galaxy 3 supercomputer one day.

Scientists used AI neural networks to teach a robot dog new tricks, enabling it to climb wooden crates and hop between them.

The new processor stores data in modified DNA molecules and uses microfluidic channels to perform basic computations.

The scientist who configured a small drone to target people with facial recognition and chase them at full speed warns we have no defenses against such weapons.

We could build an AI that demonstrates generalized, human-level intelligence within three to eight years — which may open the door to a "super intelligence" in a very short space of time.

The T-Flight is a maglev train that hit a record-breaking speed of 387 mph on a short test track — but engineers want to double that rate so the train can carry passengers at speeds faster than if they were traveling by plane.

Scientists used "knowledge distillation" to condense Stable Diffusion XL into a much leaner, more efficient AI image generation model that can run on low-cost hardware.

Say goodbye to earbuds that break down and create fuzz. New ultrasonic audio chip could lead to digital headphones with better noise-cancelling and spatial audio.

Qubits are normally made from superconducting metals and need to be cooled to near absolute zero to avoid collapsing. But scientists just built an error-free "logical qubit" from a single laser pulse — and it works at room temperature.

AI chatbots can't remember things well. However, scientists might have fixed AI's critical short-term memory issue, while OpenAI is also beginning to roll out long-term memory for ChatGPT.

Boston Dynamics has released footage showing how its flagship Atlas humanoid robot would cope in a factory environment as it lifts heavy struts and puts them into a flow cart.

Two of the three last-known surviving Q1 microcomputers have resurfaced after they were last known to have been used by an oil drilling company in the 1970s.

In a new video, 1X's EVE robots work together in silence in a test environment, performing actions such as sorting mail, handling objects and tidying up a child's toys.

Universal memory promises to replace both RAM and flash storage in computers with a better, faster and more energy-efficient alternative — and researchers have just moved this one step closer to reality.

The MadRadar hack bypasses the anti-spoofing protections in the radars of self-driving cars and can trick targets into imagining vehicles that aren't there — or hiding other ones that are.

The newly announced Spark v3.5 can beat OpenAI's GPT-4 in language workloads, among other areas, and can synthesize human speech that conveys different emotions, tones and speech patterns, its creators claim.

SIEVE is a new approach to web caching that's simpler and more effective than today's state-of-the-art algorithms, its creators claim — and big tech companies are taking notice.

Two pilotless helicopters completed a dozen test flights, while attempting to avoid more than 150 virtual aircraft, to test the systems that will power future air taxis.

QuEra has dramatically reduced the error rate in qubits — with its first commercially available machine using this technology launching with 256 physical qubits and 10 logical qubits.