All articles tagged with #on device ai
Anthropic is trialing a feature that lets Claude execute tasks directly on a user’s computer after prompts from a phone, enabling actions like opening apps, navigating browsers, and exporting PDFs with permission prompts; the move adds to the push for autonomous AI agents competing with Nvidia-backed OpenClaw, and integrates with Dispatch for ongoing task management.
Microsoft hired Ali Farhadi, the former AI2 CEO and Xnor.ai cofounder, as a corporate vice president reporting to Mustafa Suleyman, to join an expanding in-house AI team that aims to train frontier models with Microsoft data and compute and advance on-device AI over the next five years.
Apple's M5-powered MacBook Pros—M5, M5 Pro, and M5 Max—bring a major performance leap with up to a 40-core GPU, doubled memory bandwidth, and 128GB of unified memory and 2TB SSD options, plus Thunderbolt 5. They enable on-device AI with neural accelerators, letting creatives run AI models locally for faster iteration on tasks like 3D rendering, video editing, and photo culling, and support on-location workflows without cloud connectivity. The emphasis is on enabling new workflows and faster iteration rather than just chasing benchmark numbers.
Lenovo rolled out next-gen AI-optimized ThinkPad and ThinkStation workstations with dedicated NPUs, Intel/AMD CPUs, and NVIDIA RTX PRO 6000 Blackwell GPUs to accelerate on-device AI, design, CAD, and data workflows, while highlighting enhanced security and sustainability. It also unveiled the ED1000 silicon-anode battery—the world’s first 1,000 Wh/L energy density—for notebooks and workstations—promising longer runtimes and higher performance. Availability spans spring to mid-2026 across multiple models in select markets, with model-specific timelines (e.g., ThinkPad P14s/P16s and ThinkStation P5 Gen 2) announced.
Apple unveils the MacBook Neo, a $599 13-inch Mac powered by the iPhone A18 Pro chip, claiming faster everyday tasks and on-device AI, with 8GB RAM, 256GB/512GB storage, two USB-C ports, a 1080p camera, and up to 16 hours of battery; education pricing is available and pre-orders begin now with availability March 11.
AMD announced an expanded Ryzen AI 400 Series for desktops and Ryzen AI PRO 400 Series for mobile, enabling on‑device AI and Copilot+ experiences with up to 50 TOPS on desktops and 60 TOPS on mobile, plus up to 30% faster multithreaded performance on PRO notebooks and enterprise-grade AMD PRO security, with OEMs including HP and Lenovo for desktops and Dell/HP/Lenovo for mobile workstations.
Discord plans a global rollout of age verification for adult-content access using on-device facial age estimation or off-device government-ID checks, with selfies kept on the user’s device and IDs verified off-device before deletion. This follows a breach that exposed 70,000 IDs and has sparked privacy and security concerns about data handling and verification accuracy, as some users may avoid verification or find ways to bypass it while Discord contends most users will only verify once.
Reolink’s TrackFlex Floodlight WiFi and Elite floodlight cameras use on-device AI to detect people, vehicles, and pets, enabling fast search of locally stored footage without cloud processing or subscriptions. Features include dual lenses, integrated floodlights, PIR sensors, 355° horizontal pan, 50° vertical tilt, and 270° detection with a 3,000-lumen floodlight and 110 dB siren. The lineup offers panoramic 180° 4K+ (Elite Pro PoE) or 180° 4K (Elite Floodlight WiFi) options, plus microSD storage, Google Assistant support, and IP66 weatherproofing, all designed to speed responses while preserving privacy and avoiding ongoing fees.
Google researchers reveal a two-stage on-device method for extracting user intent from interactions, keeping data on the device to protect privacy and outperforming some smaller models and multimodal LLMs in tests. The approach first summarizes user actions on-device, then generates an overall intent, with potential applications in proactive assistance and memory, while noting limitations including Android/web scope and US-English testing and the need for guardrails.
Microsoft promotes a new Copilot+ PC category, defining Copilot+ PCs as Windows 11 devices with up to 16GB RAM, 256GB SSD, and a Neural Processing Unit of 40+ TOPs, claiming they’re essential for the next generation of computing and on‑device AI. The piece notes that older AI PCs may not qualify and labels the push as marketing, while some PC makers like Dell are diversifying beyond AI-centric specs amid rising RAM/GPU costs.
Apple's AI team has undergone a strategic restructuring focused on on-device AI and Foundation Models, with leadership changes including the departure of John Giannandrea. Despite some departures, the team remains large and active in research, with plans for a major AI revamp in 2026 emphasizing privacy and efficiency, aligning with Apple's broader product ecosystem.
The Galaxy S26 series will feature deeply integrated on-device AI powered by Samsung's Gauss model, enabling exclusive, privacy-focused AI features that could differentiate it from previous models and competitors, with potential implications for hardware and software strategies.
Google is launching a new cloud-based platform called Private AI Compute that enables advanced AI features on devices while maintaining user privacy, similar to Apple's Private Cloud Compute, allowing more complex AI tasks to be processed securely in the cloud without compromising data security.
Google DeepMind has introduced Gemini Robotics On-Device, a new AI model that enables robots to perform tasks locally without internet, demonstrating capabilities like unzipping and folding, and adaptable to various robot types, with a new SDK for developers.
Neural Processing Units (NPUs) are specialized processors designed to handle AI and machine learning tasks more efficiently than traditional CPUs or GPUs. Big tech companies like Google, Microsoft, and Apple are increasingly integrating NPUs into their devices to support on-device AI capabilities, promising faster and more efficient AI applications. However, the current software demand for these capabilities has not yet caught up with the hardware advancements, leading to a mix of hype and genuine innovation in the tech industry.