New Tool Detects AI Vision Hallucinations

Researchers at Los Alamos National Laboratory have developed a new tool called the Prelim Attention Score (PAS) to detect hallucinations in vision-language AI models, systems that combine image analysis with large language models. These models can sometimes generate descriptions of objects or details that are not actually present in an image. PAS works by monitoring how much the AI relies on the visual input versus its own previously generated text while producing a response, helping identify when the model is beginning to make things up.

The method operates in real time and can be integrated into existing vision-language models. By analyzing internal attention patterns, PAS provides a score indicating the likelihood of hallucination, allowing developers and users to assess the reliability of AI-generated outputs. The approach achieves state-of-the-art accuracy in detecting hallucinations and could improve the safety and trustworthiness of AI systems used in applications such as autonomous vehicles, healthcare imaging, robotics, and security monitoring.

More information:

https://interestingengineering.com/ai-robotics/us-tool-hallucinations-machine-vision-model

Digital Preservation of the Endangered Vaquita

Researchers from Florida Atlantic University have created one of the most detailed digital records ever made of the vaquita (Phocoena sinus), the world’s most endangered marine mammal. Using medical CT scans, high-resolution micro-CT imaging, and digital photography, the team digitized a rare female vaquita skeleton collected in 1966, transforming it into interactive 3D models that can be explored, measured, and studied without risking damage to the fragile original specimen. The resulting datasets have been made freely available through the MorphoSource repository.

The project serves as both a scientific resource and a conservation effort for a species on the brink of extinction. Found only in Mexico’s northern Gulf of California, the vaquita has suffered a catastrophic population decline due to accidental entanglement in illegal gillnets used to catch totoaba fish. With only a handful of individuals believed to remain in the wild, the digital archive preserves invaluable anatomical information for future generations while raising awareness of the urgent need for international action to eliminate gillnet fishing and protect this critically endangered porpoise.

More information:

https://www.fau.edu/newsdesk/articles/pixels-preserve-endangered-vaquita.php

AI Brain Implant Restores Communication and Employment for ALS Patient

Researchers at the University of California, Davis have demonstrated a major advance in brain–computer interface (BCI) technology that enabled a 47-year-old patient with amyotrophic lateral sclerosis (ALS) to communicate, control a computer, and maintain full-time employment despite losing the ability to speak. The implanted system uses AI-driven decoding algorithms to translate neural activity associated with intended speech into text and synthesized speech, while also providing cursor control for independent computer use. Over more than two years of daily use, the patient generated nearly two million words and over 183,000 sentences, achieving communication speeds of about 56 words per minute with high accuracy.

The study is considered a significant milestone because the system functioned reliably outside the laboratory and without constant researcher supervision. Researchers reported that the participant accumulated more than 3,800 hours of use, creating the largest known single-neuron-resolution brain recording dataset of its kind. Beyond restoring communication, the technology allowed the user to reconnect socially, preserve aspects of his natural voice, and continue professional work, highlighting how AI-enhanced BCIs are moving from experimental demonstrations toward practical tools that can meaningfully improve quality of life for people with severe paralysis.

More information:

https://www.theregister.com/science/2026/06/16/ai-and-brain-computer-interface-allow-speechless-als-patient-to-work-a-full-time-job/5256492

BlueME Enables Long-Range Underwater Robot Communication

Researchers at the University of Florida have developed BlueME, a novel underwater communication system that enables marine robots to exchange data directly over distances exceeding 700 meters without needing to surface. The technology is based on compact magnetoelectric antennas, which use very-low-frequency electromagnetic signals to transmit information through water more efficiently than traditional acoustic or optical communication methods. This capability allows autonomous underwater vehicles to coordinate missions, share sensor data, and make collective decisions in real time while remaining submerged.

The breakthrough could significantly enhance applications such as ocean exploration, environmental monitoring, infrastructure inspection, and search-and-rescue operations. Unlike conventional underwater communication systems that often suffer from limited range, high power consumption, or environmental interference, BlueME operates with relatively low energy requirements while maintaining reliable robot-to-robot connectivity. The research team has already demonstrated a working prototype and is seeking further funding to advance the technology toward larger-scale deployment in marine robotic networks.

More information:

https://news.ufl.edu/2026/05/marine-robot-communication/

Apple Smart Glasses Delayed to 2027

Apple’s long-rumored smart glasses project has reportedly encountered new development hurdles, pushing the expected launch from early 2027 to late 2027. The device, known as N50, is designed to compete with products such as Meta’s Ray-Ban smart glasses and is expected to focus on AI-powered assistance, cameras, audio features, and voice interaction rather than offering a full augmented reality display. Reports suggest that delays are linked in part to Apple’s ongoing efforts to upgrade Siri and integrate more advanced Apple Intelligence capabilities into the wearable experience.

Apple reportedly considers smart glasses a strategic priority and views the technology as a long-term successor to today’s mobile devices. The company is said to be experimenting with multiple frame designs and color options while aiming to differentiate itself through tight integration with the Apple ecosystem. At the same time, Apple appears to be reassessing its broader Vision product roadmap, with lighter and more affordable Vision headsets now expected no earlier than 2028–2029 as resources are redirected toward making smart glasses a mainstream consumer product.

More information:

https://www.cnet.com/tech/mobile/apple-smart-glasses-development-bumps-reported-delay/

A Robot Companion Supporting Independent Aging

Researchers in New Hampshire are testing a socially assistive robot called Robbie to help older adults remain independent at home. The robot, based on Hello Robot’s Stretch 4 platform, assists elderly by providing exercise guidance, medication and meal reminders, hydration prompts, and personal hygiene cues. Developed through a collaboration between the University of New Hampshire and the National Institute on Aging, the project aims to address growing shortages of home-care workers while supporting aging populations who wish to remain in their own homes.

Unlike the humanoid robots often portrayed in science fiction, Robbie prioritizes functionality over appearance. Equipped with cameras, sensors, and a mobile platform, it can monitor routines, deliver reminders, and assist with everyday tasks, reducing the burden on family caregivers. Although the system currently costs nearly $30,000 and remains in an early adoption phase, researchers view it as a practical example of how robotics could complement human care in the coming decades as demand for elder-support services continues to rise.

More information:

https://apnews.com/article/robot-elder-care-companion-946ce0517281381950e72f088b0eda89

Brain-Inspired AI Runs on Ambient Energy

Researchers at the University of Southern California (USC) have unveiled a miniature neuromorphic computing system that operates without batteries or external power, harvesting ambient energy from its surroundings to sense, process information, and make decisions in real time. Inspired by the human brain, the fully analog device integrates sensing, computing, and memory into a single energy-efficient architecture, enabling intelligent operation at the edge without reliance on cloud infrastructure.

The breakthrough represents a significant step toward self-sustaining artificial intelligence systems for applications such as environmental monitoring, autonomous robotics, smart infrastructure, and wearable technologies. By dramatically reducing power requirements while maintaining adaptive decision-making capabilities, the technology could pave the way for a new generation of autonomous devices capable of operating in remote or resource-constrained environments.

More information:

https://viterbischool.usc.edu/news/2026/06/usc-scientists-built-a-tiny-system-that-thinks-like-a-brain-and-runs-on-thin-air/

ViWAS Kick-Off Meeting

A new ambitious European initiative, the Virtual Worlds Academy of Skills (ViWAS), has been unveiled under the Digital Europe Programme, aiming to position Europe at the forefront of Web 4.0 and immersive technologies. Coordinated by Professor Fotis Liarokapis, CYENS Centre of Excellence, the project brings together a large, multidisciplinary consortium spanning over 16 countries to tackle one of Europe’s most pressing challenges: the growing shortage of advanced digital skills in virtual worlds and extended reality (XR). Through innovative, simulation-based learning environments and immersive digital twins, ViWAS will enable users to gain hands-on experience in safe, scalable, and cost-effective settings.

ViWAS will establish a pan-European training ecosystem designed to equip learners, professionals, and institutions with cutting-edge competencies in XR, artificial intelligence, and virtual environments. The ViWAS project started on the 1^st of May and the kick-off meeting took place in Paris, France between 26-27 May 2026. The Academy will function as an open, interoperable ecosystem, connecting universities, vocational training providers, industry, SMEs, and public institutions. Leveraging AI-driven personalization and advanced XR platforms, ViWAS will deliver tailored learning experiences, foster cross-sector collaboration, and support workforce upskilling and reskilling at scale.

More information:

https://ec.europa.eu/info/funding-tenders/opportunities/portal/screen/opportunities/projects-details/43152860/101298981/DIGITAL?order=DESC&pageNumber=1&pageSize=100&sortBy=es_SortDate&keywords=101298981