Modelos, agentes, investigación y negocio de la IA.
France becomes the first country to set concrete deadlines for adopting quantum-resistant cryptography. From 2027, ANSSI will not certify products without post-quantum encryption, and by 2030 only secure solutions against quantum attacks should be purchased.
Researchers at the University of California, Davis, have developed a brain implant that allows an ALS patient to speak with 99% accuracy, producing nearly 2 million words over two years. The study, published in Nature Medicine, marks a milestone in the duration and autonomy of brain-computer interfaces.
Roman Imankulov, a Python developer, was about to install a malicious repository disguised as a proof of concept for a job offer. His intuition led him to use an AI agent (Codex) to analyze the code, which detected a hidden backdoor in a test file. The incident reveals how attackers exploit trust in npm install and the importance of automated review.
Japanese startup Sakana AI introduces Marlin, a B2B agent that spends hours researching and synthesizing strategic reports. It abandons the instant speed of chatbots for deep, autonomous reasoning, targeting corporations, financial institutions, and think tanks.
He Tingbo, Huawei's chief semiconductor engineer, reappeared at the IEEE International Symposium on VLSI Technology to present a new scaling law called 'Tau,' which promises to improve the efficiency of AI chips without the need for advanced nodes. This move could redefine competition in the semiconductor market.
NVIDIA has released its XR AI framework in public beta, enabling developers to create multimodal agents for augmented reality glasses and XR devices. These agents process vision, audio, and language simultaneously, opening the door to hands-free contextual assistants.
Nvidia has participated in the $100 million Series A round of Hydra Host, a startup developing software to efficiently manage GPU clusters, reducing reliance on specialized hardware. The investment reveals Nvidia's strategy to maintain its influence in the AI ecosystem while the industry seeks alternatives to its costly chips.
The U.S. government has intervened in a lawsuit against xAI, defending the use of highly polluting gas turbines in its data centers. It argues that shutting them down for environmental reasons would compromise national security, in a case pitting the NAACP against Elon Musk.
According to Gizmodo data, ChatGPT's share of the AI assistant app market has fallen below 50% for the first time. The shift marks the end of an era of dominance and the beginning of more fragmented competition, with rivals like Google Gemini, Microsoft Copilot, and emerging startups gaining ground.
Qualcomm has unveiled the Snapdragon Reality Elite, its flagship chip for augmented and mixed reality glasses. With Android XR as the reference operating system, this processor aims to standardize the hardware of upcoming devices and accelerate metaverse adoption.
The U.S. government pressured Anthropic to remove Claude Fable 5 and Mythos 5, models with advanced hacking capabilities. However, analysts warn this action only delays the inevitable: the proliferation of dangerous AI will be the norm, not the exception.
Sina Weibo's VibeThinker-3B claims to outperform systems with hundreds of billions of parameters in math and code. However, the scientific community doubts: is it a true efficiency breakthrough or a symptom that benchmarks no longer measure what they should?