Infrastructure Correspondent
Adrian Cole follows the infrastructure behind the industry: chips, data centers, power and the economics of compute. His coverage asks what systems cost to build and operate, where the bottlenecks sit and which performance claims survive a like-for-like comparison.
905 published articles · Showing 41–60
The contribution, made through Nvidia's MGX 1. 2 reference architecture, extends a modular blueprint to OEMs, ODMs and cloud operators for assembling high-density accelerated computing infrastructure....
Read article →A preprint claims that DiffusionART, a new reinforcement learning algorithm, fine-tunes video diffusion models with 12 times fewer GPU hours than existing methods and 28% to 53% higher rewards, while removing a hidden instability in forward-process RL....
Read article →Microsoft’s new decision-scoring model aims to replace expensive LLM calls for high-volume classification and routing tasks, potentially lowering compute costs for enterprise applications....
Read article →OpenAI published a massive collection of machine-generated mathematical results, drawing alarm from mathematicians who told The Verge the field will need years to digest the work....
Read article →The Allen Institute for AI shifted its GPU cluster scheduling from a priority-based approach to a system of time budgets and fair-share, aiming to direct scarce GPU hours toward the highest-impact research and reduce operational toil....
Read article →The demonstration illustrates how large language models can automate the assembly of physics, rendering and sensor-simulation pipelines, potentially shrinking the manual engineering effort required to create digital twins and robotics testbeds....
Read article →The announcement follows a series of keynote bookings that Automatica has reported, including Lenovo Chairman and CEO Yuanqing Yang, Waymo co-CEO Tekedra Mawakana, and Stellantis CEO Antonio Filosa, all reflecting the show’s emphasis on AI-driven automotive and consumer…...
Read article →"CES brings together leaders from across industries to showcase technology that solves real challenges, and we're thrilled to welcome Stellantis and Antonio Filosa to share their vision for the future of mobility," CTA President and CEO Kinsey Fabrizio said in the release....
Read article →The enterprise‑only agent runs persistently in the cloud, can spawn sub‑agents and coworker identities, and is available only in a closed preview with no pricing disclosed....
Read article →Google’s Playground lowers the barrier to game creation by converting text prompts into browser-based games, though creation access is tied to AI subscription tiers....
Read article →OpenAI is expanding ChatGPT for Teens with a College Planner tool aimed at U.S. high schoolers, along with flashcards, quizzes and a note-capture feature, as it seeks to make college prep more accessible....
Read article →The European Commission’s draft guidelines for high-risk AI classification offer practical examples and a revised enforcement timetable, giving providers a first look at how the EU AI Act will be interpreted....
Read article →Microsoft’s first Nvidia-powered Surface devices—the Laptop Ultra and RTX Spark Dev Box—go on pre-order alongside Windows 11 capabilities that let AI agents run locally, a move designed to cut cloud costs and latency for coders and creators....
Read article →The CTA confirmed Waymo co-CEO Tekedra Mawakana will deliver a keynote at CES 2027 on January 6, detailing safety standards, international expansion, and fleet plans for an autonomous ride-hailing service that now claims over 500,000 trips per week....
Read article →Filosa joins a headline keynote lineup that includes Waymo Co-CEO Tekedra Mawakana and Lenovo CEO Yuanqing Yang, according to CES....
Read article →The collaboration marks a concrete step in the commercial race to field general-purpose humanoid robots by integrating LG’s component manufacturing scale with Nvidia’s physical AI computing stack....
Read article →The work trains a 50‑million‑parameter projector between the frozen vision encoder of Kimi K2. 6 and the GLM 5....
Read article →A preprint proposes a technique that adds an adapter’s weight update to a post-trained model, avoiding repeated post-training and cutting preference-coherence loss by more than half compared with common practice....
Read article →The study challenges the assumption that pinpointing the layers responsible for a model's safety behavior can yield a durable guardrail against fine-tuning attacks....
Read article →The work builds on the ProCo framework, which uses exact von Mises-Fisher (vMF) scores for probabilistic contrastive learning. The authors prove that the leading-order term in the score is \(A_c/\tau\), where \(A_c\) is the mean resultant length, a measure of class concentration....
Read article →