Topic
Explore the latest artificial intelligence news and research.
AWS has released Strands Decider 2B, a small decision model. This article compares its delivery model with Jev and OpenAI's Decisions API, and covers local inference and retraining, latency on long inputs, and how confidence scores relate to tool execution, based on official materials.
A U.S. district court dismissed Chegg's and PMC's antitrust suits over Google's AI search features. A close reading of the 41-page opinion shows how each claim fell short, and what Google's new opt-out and UK rules mean for publishers.
Security firm Glow reports that AI coding agents published more than 13,000 internal images on GitHub. We compare gitshot's default storage with private attachment authentication and the GitHub CLI feature released on September 1, and consider what to review in shared skills and publishing permissions.
Sony has announced QSSR, an AI upscaling technology for the standard PS5. In Japan, it will arrive in two games on October 2. Drawing on official explanations and independent testing, we look at how it improves image quality, what it costs in performance, and where the PS5 Pro still differs.
The FT reports Tencent signed a five-year, $7 billion deal to use about 100,000 AI chips at Oracle's Southeast Asian facilities. We look at the upfront payments in its earnings and the regulatory conditions that remain even when chips are rented overseas.
Google is replacing Gemini's Gems with Skills, which reuse instructions in a common file format. But default tools, Workspace sync and Opal migration come with limits, so existing workflows need to be checked.
Q/C Technologies has announced a research collaboration with Sandia's CINT on an optical processing unit for AI inference. Drawing on the primary announcement and a related paper, we look at the challenges of building optical computing into a working device, how CINT's user program operates, and what conditions are needed to evaluate performance.
Micron expects humanoid robots to need over 200GB of memory and several terabytes of storage, a forecast benchmarked on autonomous vehicles. We compare it with current products and look at new fab timelines and long-term contracts to see what it means for memory supply and demand.
Anthropic's AI infrastructure deals total at least $518 billion over roughly 10 years, and about 80% require payment even if usage falls, Reuters reported. We cross-check official capacity plans and funding to see how supplier diversification differs from payment obligations.
A study comparing 27 AI models across 155 topics found that chatbot answers contain less diverse knowledge than web search results. We look at what the 18.7% figure actually compares, how much RAG helps, and the bias toward English-language sources.
