Google launched Flow, an AI-powered video tool for filmmaking using Veo, Imagen, and Gemini models. It offers features like camera control, scene building, asset management, and Flow TV for creative insights. Initially available in the U.S. via Google AI Pro and Ultra plans, Pro users get 100 generations monthly, while Ultra users enjoy higher limits and access to latest models.

Google expanded Project Mariner, its experimental AI agent, allowing simultaneous handling of up to ten tasks. Accessible via the $249.99/month AI Ultra plan, it integrates with Gemini API and Vertex AI for developers. Running on cloud-based virtual machines improves multitasking capabilities, surpassing earlier browser-based limitations. Project Mariner competes with OpenAI's Operator and Amazon's Nova Act while partnering with services like Ticketmaster and StubHub for automation. Additionally

Google unveiled Android XR, a new OS for VR/AR devices integrating with its Gemini AI. Initially for VR headsets, it will expand to AR glasses with microphones, speakers, and cameras. These glasses work with smartphones, enabling app access, real-time translation, and AI-driven features like navigation and context understanding. Partnerships with brands like Gentle Monster and Warby Parker ensure stylish designs, while Prescription lenses will be supported. Chinese firm Xreal Technology also ann

Google DeepMind launched SynthID, a tool detecting AI-generated content across images, audio, video, and text by identifying watermarks embedded during creation. Initially for images, it now includes other formats, aiding transparency in AI-generated content. SynthID subtly alters word probabilities in text without affecting meaning, watermarking outputs from Google's AI models like Gemini. Available in limited release, it targets journalists, educators, and researchers. This comes amid rising c

Microsoft is enhancing Edge with new AI APIs, enabling developers to integrate AI features via built-in models like Phi 4 mini, which has 3.8 billion parameters and excels in math. The APIs include writing assistance tools and an upcoming translation API, promoting cross-platform standards and on-device processing for privacy. Available in Edge Canary and Dev channels, Microsoft also adds a PDF translation tool supporting over 70 languages, rolling out widely next month.

Microsoft launched Windows AI Foundry at the Build Conference, replacing Copilot Runtime. This platform supports AI development from model selection to deployment, running models locally with hardware optimization for CPUs, GPUs, and NPUs. It includes APIs for common AI tasks, LoRA support for customization, semantic search tools, and local deployment via Foundry Local. Microsoft collaborates with AMD, Intel, Nvidia, and Qualcomm to ensure seamless performance across devices, promoting an open a

Microsoft introduced a new GitHub Copilot feature enabling it to function as an AI agent, performing tasks like code generation and testing with minimal supervision. Powered by the Model Context Protocol (MCP), it integrates multimodal capabilities and operates within GitHub Actions. The update includes Agent-Based DevOps for automating software lifecycle stages and an open-source initiative for GitHub Copilot in VS Code. Additionally, Microsoft launched a Site Reliability Engineering agent for

Microsoft launched NLWeb, an open-source project transforming websites into AI-powered apps using natural language capabilities. Announced at Microsoft Build 2025, it simplifies creating interfaces embeddable in any site. NLWeb uses semi-structured data and large language models to enable voice queries, aiming to make every website AI-searchable. Developed by R.V. Guha, it supports multiple operating systems and integrates with various AI models and databases. Already adopted by sites like Tripa

Google has launched the NotebookLM Android app ahead of schedule, expanding its AI-powered note-taking service from desktop to mobile. The app features smart summaries, document Q&A, and an audio overview function for source materials, supporting offline playback. Users can create new notebooks, upload content, and access existing notes with both light and dark modes available. The iOS version is set for release on Tuesday.

Microsoft now offers managed services for Grok AI models via Azure AI Foundry, marking it as one of the first hyperscale cloud providers to do so. Grok, developed by Elon Musk's xAI, is known for its unfiltered approach and ability to tackle controversial topics. The Azure versions of Grok 3 and Grok 3 mini are more restricted than those on Musk's X platform but include enhanced data integration and governance features. Controversies surrounding Grok include image editing capabilities and unauth

NVIDIA unveiled an expanded robotics development infrastructure at COMPUTEX, including the GR00T N1.5 humanoid robot model. Created in 36 hours using synthetic data, it addresses the data collection bottleneck. Tools like GR00T-Dreams generate motion data for training, enhancing robot adaptability and task success rates. Major companies like Boston Dynamics and Foxconn are adopting NVIDIA's platform, with potential industrial impacts driving what CEO Jensen Huang calls "the next industrial revol

Microsoft's Healthcare Agent Orchestrator uses AI to streamline cancer care planning by analyzing medical data, clinical trials, and records. Stanford Medicine is implementing the system to accelerate tumor board preparations, reducing administrative burdens. Integrated with Microsoft 365 tools, it coordinates specialized AI agents for radiology, pathology, genomics, and more. Partners like Johns Hopkins and UW Health are refining its real-time clinical applications, aiming to cut review times f

OpenAI launches Codex, a cloud-based AI tool integrated into ChatGPT Professional, Team, and Enterprise versions. It automates coding tasks like writing, debugging, and testing within secure sandboxes. Powered by codex-1, it uses reinforcement learning to generate human-like code and supports AGENTS.md files for project guidance. Codex operates in isolated containers without UI execution, requiring final validation by developers. OpenAI also introduces Codex mini, a faster, lower-latency model p

NVIDIA unveiled AI advancements at Computex 2025, including an AI marketplace and NVIDIA NVLink for semi-customized AI infrastructures. DGX Cloud Lepton connects developers with global GPU computing resources. Isaac GR00T N1.5 enhances humanoid robotics capabilities, while GR00T-Dreams generates synthetic data for robot training. NVLink Fusion scales AI factories with custom CPUs and accelerators, enabling integration with GPUs for high-performance computing.

OpenAI launches "OpenAI for Countries" under its Stargate project to help nations build democratic AI infrastructure. This follows a $500 billion US investment plan and includes constructing a supercomputing campus in Texas. The initiative aims for global economic growth, customizing AI services per country, ensuring security, and addressing environmental concerns despite current carbon-intensive operations.

Cloudflare launched a Vite plugin (v1.0) to simplify web app development on Cloudflare Workers. It integrates the Workers runtime into Vite's build process, supports React Router v7, and leverages Vite 6's environment API for consistent dev and prod environments. The plugin enhances SPAs with frameworks like React, Vue, and Svelte, streamlines Worker configuration, and supports Cloudflare's developer platform features such as KV, D1, and Workers AI. Developers can use create-cloudflare CLI for s

Robots are increasingly taking over human tasks across industries, from construction and policing to factories and space exploration. In construction, robotics automates bricklaying and welding, addressing labor shortages. Police departments deploy AI-powered robots for surveillance and incident response. Factories use advanced robots like Tesla's Optimus for logistics. Hyundai introduced service robots for office tasks, while SpaceX explores humanoid robots for Mars missions. This shift challen

AWS now offers Meta's latest Llama 4 models, Scout and Maverick, via Amazon Bedrock and SageMaker JumpStart. These multimodal models use a Mixture of Experts architecture, with Scout optimized for single NVIDIA H100 GPUs and Maverick excelling in reasoning and coding. Scout supports up to 10 million tokens, while Maverick supports 1 million tokens, surpassing Llama 3's 128K limit. Both models are available on various cloud platforms, including Databricks and Hugging Face. The Llama 4 family also

Mistral AI launched Mistral Medium 3, a cost-effective mid-sized language model for businesses. It performs comparably to larger models like Claude Sonnet 3.7, achieving over 90% of their scores at lower costs ($0.40 per million input tokens). The model supports flexible deployment options, including hybrid and local setups with as few as four GPUs, and excels in coding, customer support automation, and technical data analysis. While receiving enterprise support, it faces community concerns abou

LLM agents are AI systems that autonomously achieve objectives by breaking tasks into steps, using knowledge bases or APIs, and integrating information. They surpass single chatbots by automating complex workflows like travel booking or report generation. Analogous to digital assistants with plugin access, agents use internal reasoning and external tools. Frameworks like Agent-to-Agent (A2A) and Multi-Component Prompting (MCP) enhance agent capabilities by enabling collaboration and unified tool