ACM TechNews, Monday, November 4, 2024

Welcome to the November 4, 2024 edition of ACM TechNews, providing timely information for computer professionals three times a week.

Supercomputer Built to Transform a Country, Paid for by Ozempic

Denmark’s national AI supercomputer was unveiled recently, built with technology from Nvidia and funding from the Novo Nordisk Foundation, the largest shareholder in Novo Nordisk, maker of the blockbuster weight-loss drugs Ozempic and Wegovy. When fully operational, it will be available to entrepreneurs, academics, and scientists inside companies like Novo Nordisk, which stands to benefit from its help with drug discovery, protein design, and digital biology.
[ » Read full article *May Require Paid Registration ]

Textbooks Come Alive

Computer scientists led by Ryo Suzuki at the University of Colorado Boulder re-imagined physics textbooks so students can record a diagram on iPads to adjust settings and see for themselves the impact the changes have. The Augmented Physics tool relies on a Meta model called Segment Anything, a computer visualization tool that allows users to isolate particular objects in a photo. Students can select various objects inside a diagram and assign those objects roles; AI then applies some basic physics to make those objects move.
[ » Read full article ]

Chinese Researchers use Meta's LLM to Build a Model for Military Use

Chinese research institutions linked to the People's Liberation Army used Meta's Llama large language model (LLM) to develop an AI tool for potential military applications. The researchers added their own parameters to Meta's Llama 13B, an earlier version of the LLM, to build ChatBIT, an AI tool that can collect and process intelligence and produce reliable information for operational decision-making.
[ » Read full article ]

Bay Area Picked to Host U.S. Semiconductor Tech Center

The San Francisco Bay Area will host the National Semiconductor Technology Center, to be built with funding from the CHIPS and Science Act. One of three CHIPS for America research and design facilities, the facility is expected to generate more than $1 billion in research funding, according to the office of California Governor Gavin Newsom.
[ » Read full article *May Require Paid Registration ]

Zero-Click Flaw Exposes Popular Storage Devices to Attack

A zero-click vulnerability uncovered by researchers at security consultancy Midnight Blue in the Netherlands affects a photo application installed by default on popular network-attached storage devices made by Taiwan’s Synology. Discovered as part of the recent Pwn2Own hacking contest in Ireland, the bug would allow attackers to gain access to the devices to steal personal and corporate files, plant a backdoor, or infect the systems with ransomware.
[ » Read full article *May Require Paid Registration ]

Voting Rights Groups Concerned Chatbots Produce Election Falsehoods in Spanish

An analysis by two nonprofit newsrooms working with the Science, Technology and Social Values Lab at New Jersey’s Institute for Advanced Study found that AI chatbots generate more false claims about voting rights in Spanish than they do in English, in the lead-up to the U.S. presidential election. Assessing responses by Meta's Llama 3, Anthropic's Claude, and Google's Gemini to specific election-rated prompts, the researchers found they produced incorrect information in more than half their responses in Spanish.
[ » Read full article ]

3D Image Reconstruction to Preserve Cultural Heritage

A neural network developed by a multinational research team led by Satoshi Tanaka from Japan's Ritsumeikan University allows for the 3D reconstruction and digital preservation of sculpted and carved reliefs using old photos. The neural network performs semantic segmentation, depth estimation, and soft-edge detection, which together enhance the accuracy of 3D reconstruction. The core strength of the network lies in its depth estimation, achieved through a novel soft-edge detector and an edge matching module.
[ » Read full article ]

AI Rests on Billions of Tons of Concrete

The amount of concrete used in datacenter construction is challenging tech companies' commitments to eliminate carbon emissions and bolster demand for green concrete. In response, an Open Compute Project Foundation-led initiative to speed testing and deployment of low-carbon concrete in datacenters has garnered support from Amazon, Google, Meta, and Microsoft.
[ » Read full article ]

Musk says X Users Fight Falsehoods. The Falsehoods Are Winning

The nonprofit Center for Countering Digital Hate found that X's Community Notes has failed to combat misinformation. The crowdsourcing program allows contributors to propose notes that debunk or provide additional context to posts; only notes receiving consensus among participants are displayed publicly. An analysis of 283 posts containing false or misleading election claims found that proposed Community Notes on 229 of the posts provided accurate context, but consensus was reached and notes were publicly attached to just 20 of those posts.
[ » Read full article *May Require Paid Registration ]

14-Year-Old Named America's Top Young Scientist

Ninth-grader Sirish Subash, 14, from Snellville, GA, won the 3M Young Scientist Challenge with a device that detects pesticide residue on produce. The PestiSCAND device, tested on more than 12,000 samples of produce, measures the wavelengths of light that bounce back when directed at the surface of fruits and vegetables, and identifies specific patterns in the reflected light that are created by pesticide residue.
[ » Read full article ]

Eavesdropping on Phone Calls by Sensing Vibrations

Suryoday Basak at Pennsylvania State University and colleagues used a commercially available millimeter-wave sensor to pick up the tiny vibrations of a Samsung Galaxy S20 earpiece speaker playing audio clips. The team converted the signal to audio and passed it through an AI speech recognition model, which transcribed the speech. The system achieved a word accuracy rate of 50% and a character accuracy rate of 67%.
[ » Read full article ]

Neural Networks on the Edge

Researchers at Japan's Tokyo University of Science developed a binarized neural network (BNN) to allow for more efficient AI implementation in Internet of Things edge devices and other resource-limited devices. The researchers reduced circuit size and power consumption through the use of a magnetic random access memory (MRAM)-based computing-in-memory architecture. This required the creation of a new XNOR logic gate as the foundation for a MRAM array, which stores information in its magnetization state using a magnetic tunnel junction.
[ » Read full article ]