Modelos, agentes, investigación y negocio de la IA.
A custom PC with AMD Ryzen AI Max+ 395 and 128 GB of RAM has run a 397 billion parameter language model, a task that a year ago required servers with multiple GPUs. This milestone marks a before and after in edge AI computing.
The U.S. Federal Energy Regulatory Commission (FERC) approved a regulatory framework that streamlines the interconnection of large consumers, such as AI data centers and advanced manufacturing plants. The measure aims to reduce bureaucracy, incentivize private investment in infrastructure, and keep electricity rates affordable.
Anthropic has added Artifacts to Claude Code, allowing enterprise teams to create interactive, shareable dashboards in real time from the command line. The feature, available on Team and Enterprise plans, competes with OpenAI Codex Sites, though with a lighter, simplicity-focused approach.
AWS has announced the general availability of EC2 G7 instances, the first in the public cloud to feature NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs. They offer up to 4.6x better AI inference performance and 2.1x better graphics performance compared to G6, with significant improvements in memory, networking, and video encoding.
MIT researchers have adapted principles from nuclear reactor cooling to create a bubble system that drastically reduces energy and water consumption in AI data centers. This innovation could alleviate the sector's growing energy crisis.
France has set a pioneering mandate: by 2027, operators of critical infrastructure must migrate to post-quantum cryptography. This measure, the first of its kind at a national level, aims to stay ahead of quantum computers' ability to break current cryptographic systems.
In an unusual move, the U.S. Department of Justice intervened to defend xAI in an environmental lawsuit filed by the NAACP, claiming the company is 'vital' for national security and military operations, including the war in Iran. The case highlights tensions between technological innovation, environmental justice, and geopolitical interests.
The Pentagon has multiplied by 18 the users of its generative AI platform GenAI.mil in just six months, reaching 1.5 million daily users. This represents nearly half of the Department of Defense's workforce. The rapid adoption reflects a strategic shift toward integrating AI into military operations.
A new open standard, backed by Microsoft and Google, will allow AI agents to discover and use tools on the fly, transforming enterprise automation. The protocol, similar to a 'search engine' for agents, promises an interoperable ecosystem where AI assistants can find and execute APIs, scripts, and services in real time.
Hugging Face has released a benchmark evaluating open language models on custom tool use tasks. Results show improvement but still lag behind proprietary models in complex agent scenarios.
OVHcloud, a European cloud provider, is venturing into frontier AI model development with plans to open-source them. The move aims to position Europe as a viable alternative to US and Chinese giants amid growing data sovereignty concerns.
The adoption of artificial intelligence in the business world faces a growing obstacle: lack of transparency. Organizations demand explainable models and governance to mitigate regulatory and reputational risks, thus accelerating trust in these technologies.