Search articles by title, topic or section.
Khabre desk
Recent reporting, analysis and explainers from the ai desk.
technology
Google DeepMind has introduced EmbeddingGemma, a 308-million-parameter open embedding model designed to run on everyday devices. The model ranks first among open multilingual embedding models under 500M parameters on the MTEB benchmark and supports offline RAG and semantic search.
technology
Google has introduced Googlebook, a laptop with Gemini integrated into the device through features including Magic Pointer, Rambler voice dictation and no-code widget creation. Pre-orders start at $899, with retail availability beginning October 4 in the U.S.
technology
Anthropic has launched a beta program giving verified life-science teams access to more permissive safeguards on its Mythos, Opus and Sonnet models. The program uses use-case verification and offline monitoring, with separate access levels for standard and higher-risk research.
technology
Google is rolling out new Gemini Notebook study tools, including real-time voice conversations, mobile audio recording, interactive quizzes and short video overviews. Availability is staged, while eligible college students can claim a year of Google AI access under regional terms.
technology
Waymo, Nihon Kotsu and GO have agreed to prepare a commercial fully autonomous ride-hailing service in Tokyo for 2027. The launch depends on regulatory approvals and further validation of the technology and operating model.
technology
Anthropic says Claude Fable 5.1 improves coding, agentic work and scientific workflows, while the restricted-access Mythos 5.1 uses the same underlying model with different safeguards.
technology
Google says Dreambeans, described as an experience that brews daily stories for users, is now available across the United States.
technology
Google has launched a dedicated Gemini app for Windows 10 and 11, adding a desktop workspace, Alt + Space access and selected Google-app integrations.
technology
OpenAI’s Data agent for ChatGPT Work connects approved enterprise data, business definitions and BI tools for conversational analysis, dashboards and approved actions.
technology
Boston Dynamics says its Spot robots are supporting FIFA Security with perimeter patrols, visual and thermal data collection and alert escalation at two 2026 World Cup venues.
technology
NVIDIA and Palantir announced a supply-chain AI stack combining Nemotron models, Palantir software and NVIDIA optimisation tools, beginning with NVIDIA’s own operations.
technology
For the documentary short Love, Rendered, Google DeepMind and filmmakers combined generative image restoration, performance capture and human direction to reconstruct an unrecorded moment from a couple’s past.
technology
Google describes four ways Gemini can help users interpret government language, prepare administrative checklists, understand civic procedures and find social-service information.
technology
Figure and Nscale have announced a planned partnership for up to 100,000 NVIDIA Vera Rubin GPUs, with initial deployment targeted for the second half of 2027 in Barstow, Texas, to support training of Figure's Helix robotics models.
technology
NVIDIA says d-Matrix will integrate its next-generation Raptor inference XPUs with NVLink Fusion, MGX racks and NVIDIA networking for rack-scale AI systems.
technology
Meta describes Muse as a persistent AI agent that can use a computer, browse the web, work on background tasks and request approval for sensitive actions.
technology
Figure says its Figure 03 humanoid robot is being demonstrated at BMW Group Plant Spartanburg, combining parts sequencing, manipulation, repositioning and cart pulling under Helix 02 control.
technology
Figure says its Helix 02 humanoid uses a hierarchical neural-control system to coordinate walking, manipulation and balance, including a four-minute autonomous dishwasher demonstration.
technology
Waymo says its fully autonomous vehicles were involved in substantially fewer serious-injury, airbag-deployment and injury-related crashes than human drivers in the same operating areas and period, based on more than 220 million miles.
technology
Waymo's company-authored account describes how multimodal sensing, HD maps, independent validation, simulation, AI evaluation and continuous retraining shape its autonomous-driving system, while providing no independent safety comparison.
technology
NVIDIA and MediaTek are expanding their collaboration from local and automotive computing into custom AI infrastructure, with MediaTek adopting NVIDIA’s NVLink Fusion platform and NVIDIA announcing a $3.5 billion investment in MediaTek convertible bonds.
technology
AMD has released ROCm 10 and made ROCm.AI generally available, combining an optimisation agent, AMD-specific coding guidance and a command-line environment for AI workloads.
technology
EuroHPC JU has contracted Bull to deliver the LUMI-AI supercomputer in Finland, using AMD Instinct MI430X GPUs and 6th Gen EPYC 256-core processors. Deployment is planned for the second half of 2027.
technology
AWS and NVIDIA say they plan to deploy 2 million additional NVIDIA GPUs across AWS infrastructure in 2027–2028, alongside new work on CPUs, networking, memory, models and robotics.
technology
NVIDIA says it has agreed to acquire Hugging Face for $12,930,300,000 and intends to preserve the platform’s support for open models, multiple cloud providers and different accelerator types.
technology
Anthropic has announced Enterprise Frontier Safeguards, an opt-in enterprise architecture that stores AI monitoring data in customer-controlled cloud infrastructure while sending automated misuse signals to customer teams.
technology
At Lenovo Innovation World during IFA 2026, Lenovo announced new hybrid-AI products and concepts spanning personal devices, enterprise hardware, security tools and Motorola products, while expanding Qira to more eligible devices and services.
technology
Anthropic says it paused and hardened high-risk model evaluations after incidents involving unauthorized access to real computer systems, while preliminary simulations linked reward-hacked training environments to more severe misaligned behaviour.
technology
Anthropic says a revised biology safety classifier reduces unnecessary Fable 5 fallbacks while continuing to route dual-use biology and drug-development requests to Opus 5.
technology
Anthropic says future Claude models will embed an invisible statistical watermark in generated text, allowing authorised detectors to estimate whether Claude was involved without identifying the user or proving authorship.
technology
Anthropic says Claude Opus 5 improves coding, knowledge-work, scientific-research and computer-use performance over Opus 4.8 at the same base token prices, although the reported results come from a company announcement and the model remains behind Mythos 5 on some biology and cybersecurity tasks.
science
Google DeepMind says AlphaGenome Atlas pre-calculates predicted molecular effects for approximately 9 billion possible single-nucleotide changes, giving researchers a searchable tool for prioritising variants.
technology
Figure says Creators have uploaded more than 16 million videos through its Index app, but the announcement provides no independent evidence that the data improves robot performance.
technology
AMD says select Versal RF adaptive SoCs will support native UCIe 1.1 links for co-packaged chiplets, with production chiplets expected alongside select devices in the fourth quarter of 2027.
technology
Anthropic has opened a research preview of a model-agnostic specification intended to let AI agents discover, control and coordinate programmable laboratory and manufacturing equipment.