Apple just opened up its Apple Intelligence suite to public beta testers across iOS 18.1, iPadOS 18.1, and macOS Sequoia 15.1. The initial drop brings system-wide Writing Tools, a new glowing interface for Siri, and Clean Up in Photos. It is currently limited to U.S. English and newer hardware like the iPhone 15 Pro or iPhone 16 series.
Apple just rolled out iOS 18 to the public, but its biggest headline feature, Apple Intelligence, is noticeably missing from day one. Instead, users get a massive Control Center overhaul, freeform home screen icon placement, RCS messaging, and real-time Voice Memos transcription. Why it matters: Even without the new AI smarts, it is one of the most flexible iPhone updates in years.
Tesla is reportedly hosting its October 10 Robotaxi unveiling at the Warner Bros. studio lot in Burbank, California. The controlled movie set features simulated suburban homes and downtown streets, allowing Tesla to pre-map the route and avoid real-world traffic during live demos. Staging the debut on a private lot will guarantee a smooth show, but critics note it skirts the real test of chaotic public roads.
Apple just rolled out its Clean Up tool in the latest iOS 18.1, iPadOS 18.1, and macOS 15.1 developer betas. Powered by Apple Intelligence, it lets you tap, circle, or brush away photobombers and clutter while automatically filling in backgrounds and clearing shadows. For now, it is limited to devices set to English and the U.S. region.
Alibaba just released Qwen2-VL, open-sourcing the 2B and 7B models under Apache 2.0 while offering its flagship 72B version via API. The model handles dynamic resolutions, parses 20-minute videos, and powers mobile device agents. Why it matters: The 72B flagship rivals GPT-4o on visual benchmarks, bringing top-tier multimodal power to open weights and developers.
Apple just announced its next big hardware event for September 9th with the tagline "It's Glowtime." While the iPhone 16 lineup brings design tweaks and new capture buttons, the real spotlight is on Apple Intelligence, which is expected to expand across the entire flagship lineup instead of staying Pro-exclusive.
Microsoft just announced that its controversial Recall feature won't reach Windows Insiders until October, pushing back its testing timeline again. The tool—which snapshots PC activity for local search—was pulled in June after researchers exposed critical security flaws. Microsoft is currently overhauling it with default opt-in, full database encryption, and Windows Hello authentication, making a wide public release this year very unlikely.
McAfee just launched its Deepfake Detector on Lenovo's Copilot-Plus PCs to warn users of AI-generated content in real time. It processes audio entirely on-device to protect privacy and pops up alerts for suspicious media. Why it matters: It treats deepfakes like modern malware, starting with a 30-day free trial before charging $9.99 for the first year.
Microsoft just launched three new Phi-3.5 lightweight models under a permissive MIT license, featuring a 3.8B Mini, a 41.9B MoE, and a 4.15B Vision model. All three support a 128k context window and beat larger competitors on select reasoning benchmarks. It's a major gift for developers wanting capable local AI without commercial restrictions.
Qualcomm just unveiled the Snapdragon 7s Gen 3, a processor designed to bring generative AI features like on-device translation to $500 to $600 smartphones. Xiaomi will debut the chip next month, trickling down features previously reserved for $800+ flagships.
xMEMS Labs just unveiled the XMC-2400 µCooling chip, an all-silicon active micro-fan that measures just 1mm thin and weighs under 150mg. Designed to stop phones and tablets from throttling during heavy on-device AI workloads, it moves 39 cubic centimeters of air per second. Samples are slated for early 2025.
Nvidia just introduced its first on-device small language model, Nemotron-4 4B Instruct, to its ACE digital human suite at Gamescom. Distilled from a 15B model, it is optimized to run locally on RTX GPUs with minimal memory and fast response times. Why it matters: Gamers can now talk to dynamic NPCs in games like Mecha BREAK entirely on local hardware, without relying on cloud servers.
Primate Labs just launched Geekbench AI 1.0, bringing standardized performance testing to machine learning workloads across Android, Linux, macOS, and Windows. Serving as the successor to Geekbench ML, the tool gives engineers and enthusiasts a unified way to see how their hardware actually handles AI tasks.
Google just introduced 4 new AI-powered accessibility upgrades across Pixel and Android devices. Pixel's Guided Frame now lives directly in camera settings with better subject focus, while the Magnifier app can search for real-world text on menus or signs. Foldables like the Pixel 9 Pro Fold gain a dual-screen Live Transcribe mode, alongside 7 new languages for Live Caption.
Google just launched the Pixel 9 series, powered by the custom Tensor G4 chip and up to 16GB of RAM. The lineup is the first to run multimodal Gemini Nano on-device, unlocking generative tools like Pixel Studio, Pixel Screenshots, and conversational Gemini Live.
Apple is reportedly delaying Apple Intelligence past the initial September release of iOS 18, pushing public rollout to an October software update. That means the first batch of iPhone 16 buyers won't get AI out of the box, as Apple buys more time to squash bugs. Developers are expected to get early access via an iOS 18.1 beta this week.
AMD just teamed up with TensorStack to release Amuse 2.0 Beta, a local AI suite that runs Stable Diffusion and ControlNet workflows without any command-line setup. The biggest perk is XDNA Super Resolution, which speeds up creation and doubles output resolution on supported Ryzen AI processors and Radeon RX 7000 GPUs. It gives PC users a seamless on-device sandbox that bypasses the cloud entirely.
Apple's research team just released open DCLM models with 7B and 1.4B parameters on Hugging Face, sharing weights, code, and training datasets. The 7B model scored 63.7% on MMLU, outperforming Mistral-7B while using 40% less compute. It shows Apple's data curation prowess, though it's research-focused rather than on-device software.
Meta just unveiled MobileLLM, a specialized architecture designed to squeeze capable language models under 1 billion parameters onto smartphones. By prioritizing network depth over width and sharing weights across blocks, its tiny 350M version matches the 7B LLaMA-2 on specific API tasks. Meta has open-sourced the pre-training code, setting up a new path for fast on-device assistants.
French research lab Kyutai just unveiled Moshi, an open-source conversational voice AI boasting an end-to-end latency of only 200 milliseconds. Built from scratch in six months, it can listen and talk at the same time and run directly on a laptop without an internet connection.
