Nvidia researchers have developed a new technique that dramatically reduces the memory demands of large language models (LLMs), potentially revolutionizing AI reasoning and lowering infrastructure costs, according to a report from VentureBeat. This innovation, called dynamic memory sparsification (DMS), compresses the key value cache, allowing LLMs to "think" longer and explore more solutions without sacrificing speed.
The development of DMS addresses a significant bottleneck in LLM reasoning, VentureBeat reported. This technology allows for an eight-fold reduction in LLM costs.
In other news, the MIT Technology Review newsletter highlighted an interview with US Deputy Health Secretary Jim O'Neill, who is promoting longevity research while defending controversial vaccine schedule changes.
Meanwhile, in the realm of archival storage, a review of M-Disc archival capability was published on Hacker News. The review, conducted by Mol Smith UK in September 2016, examined the long-term testing results of M-Discs. The author noted the challenges of long-term data preservation, citing the limited lifespan of computer drives and the potential for data loss due to magnetic creep and chemical migration. "Anyone who wishes to archive material...is faced with an almost unsolvable problem when one considers our over-reliance on new technology," Smith wrote.
In a separate development, a study suggesting moderate caffeine intake might reduce dementia risk was highlighted in Nature News. The study, which involved 130,000 people, found a link between coffee consumption and slower brain aging. Nature also reported on the use of AI to decode the rules of a long-forgotten ancient Roman board game.
Finally, in sports news, Vladyslav Heraskevych of Team Ukraine participated in the Men's Training Heat 3 at the Milano Cortina 2026 Winter Olympic games, according to NPR News. The event took place on February 10, 2026, at the Cortina Sliding Centre in Cortina d'Ampezzo, Italy.
Discussion
AI Experts & Community
Be the first to comment