The latest generative AI models such as OpenAI’s ChatGPT-4 and Google’s Gemini 2.5 require not only high memory bandwidth but also large memory capacity. This is why generative AI cloud operating companies like Microsoft and Google purchase hundreds of thousands of NVIDIA GPUs.

As a solution to address the core challenges of building such high-performance AI infrastructure, Korean researchers have succeeded in developing an NPU (neural processing unit) core technology that improves the inference performance of generative AI models by an average of more than 60% while consuming approximately 44% less power compared to the latest GPUs.

Professor Jongse Park’s research team from KAIST School of Computing, in collaboration with HyperAccel Inc., developed a high-performance, low-power NPU core technology specialized for generative AI clouds like ChatGPT.

Rene Turcios, 29, from San Francisco, has won 200 IT hackathons in two years, an ambitious achievement for someone with no programming skills.

René is a professional Yu-Gi-Oh! player, cannabis enthusiast, and reseller of Labubu toys, but he devotes most of his time to participating in IT hackathons. Since 2023, he has attended more than 200 events and won cash prizes and software credits.

The craziest thing about all this is that Tursios has no programming skills and is a representative of a new generation of programmers — the so-called web coders, i.e. people who write code with the help of AI chatbots.

You’re not bad at math. You’ve just not been zapped enough.

A group of researchers at the VIB‑UGent Center for Medical Biotechnology has developed a new platform to isolate and analyze extracellular vesicles (EVs), nanosized particles secreted by cells and playing a role in cellular communication and disease development. Called FAEVEr, the method increases the throughput of EV enrichment and is significantly more cost‑efficient than existing methods. The study is published in the Journal of Extracellular Vesicles.

Extracellular vesicles (EVs) are small particles that carry proteins, RNA, and other biomolecules from their cell of origin. They hold much promise for diagnostics and therapeutics, but isolating them from complex biofluids at high purity and throughput remains a major challenge. EVs are incredibly small—between 30 and 150 nanometers in size.

To capture these tiny containers of messengers, scientists need to rely on sophisticated equipment such as ultracentrifuges. Unfortunately, these traditional methods of EV enrichment are time‑consuming and resource‑intensive with relatively low throughput.

Our metabolic processes differ depending on the time of day and many of them are more active in the morning than in the evening. Although studies show that eating late in the day is associated with an increased risk of obesity and cardiovascular diseases, little is known about how the time we eat affects glucose metabolism and to what extent this is genetically defined.

Prof. Olga Ramich from the German Institute of Human Nutrition Potsdam-Rehbrücke (DIfE) and her team recently investigated this using data from a twin cohort from 2009-10. Their article was published in the journal eBioMedicine.

The circadian system is a hierarchically structured 24-hour time control system in the body that regulates behavior and metabolism via a central clock in the brain and peripheral clocks in organs such as the liver or pancreas. As a result, our metabolic processes differ depending on the time when we eat, which leads to diurnal fluctuations in glucose metabolism and the release of hormones after a meal.

A team of researchers from Utrecht University, Durham University, and other institutions have observed something remarkable at a chimpanzee sanctuary in Zambia. Several chimpanzees from one particular group were seen dangling blades of grass from their ear holes or their behinds, for no apparent reason. The behavior was not seen in other chimpanzee groups at the same sanctuary, despite similar living conditions.

Kits sold directly to consumers to check a variety of health metrics provide little value when it comes to guiding health decisions.

That’s big news for the most mysterious phase of matter—and maybe physics as we know it.

New research shows that retinal neurons can rewire to preserve vision in retinitis pigmentosa, a genetic disease causing blindness.

In recent years, the analysis of single-cell and spatial data has revolutionized biomedical research, making it possible to observe what happens in biological samples with an unprecedented level of detail. Interpreting this data, however, is not easy because different software offers different results which are hard to compare.

Taking this issue as the starting point, a research group from the University of Trento has developed the “Cell Marker Accordion,” a bioinformatics tool that makes the identification of cell types in the new generation data clearer and more robust. The results of the research, conducted in collaboration with Yale University (United States), the University of Trondheim (Norway), Policlinico di Milano and the Institute of Biophysics of the National Research Council—CNR, are published in Nature Communications.

“With Cell Marker Accordion we wanted to build a tool that helps researchers not only to classify cells, but also to understand why they have been classified in a certain way,” explains Emma Busarello, a Ph.D. candidate in biomolecular sciences at the University of Trento and first author of the work.