Google's AI-powered note-taking tool, NotebookLM, is set to launch a mobile app, expanding beyond its desktop platform. The tool extracts key insights from uploaded documents and can convert results into AI-generated podcasts. This mobile expansion aims to enhance convenience for on-the-go users, retain desktop features with mobile optimization, and increase its user base. The exact release date is yet to be announced.

Google rolled out Gemini Live's real-time camera Q&A and screen sharing on Pixel 9 and Samsung Galaxy S25. Users can interact with AI via live video or screen content for instant insights. The update will expand to more Android devices through a Gemini Advanced subscription. Initially announced as "Project Astra," it supports 45 languages, is age-restricted, and available in select regions.

Meta's Maverick AI model ranks second in LM Arena tests, but concerns arise as the tested version differs from the public one. Meta labels the LM Arena variant as "experimental" and optimized for conversations. Critics question benchmark reliability, fearing tailored models mislead about real-world performance. Significant differences were noted, with the arena version using more emojis and verbose responses. Meta and Chatbot Arena are urged to clarify.

Microsoft's Azure Foundry update introduces new tools to accelerate AI app development, including pre-packaged AI models and the AI Red Teaming Agent for security testing. The service now offers enhanced features like agent evaluation tools and a Visual Studio Code extension for seamless cloud deployment, boosting developer productivity and ensuring reliable AI applications. Microsoft underscores its commitment to maintaining stable APIs for AI development.

Microsoft marks its 50th anniversary with a major Copilot update, integrating OpenAI models and introducing features like memory functions, personalization, web-based operations, podcast creation, and advanced research tools. The memory function allows Copilot to recall user preferences, while new Actions enable task execution via browsers. Copilot Vision expands to Windows and mobile, analyzing content from apps and images. Deep research capabilities integrate with Bing for complex projects. Th

Microsoft upgraded Copilot to automate online tasks like travel booking, dining reservations, and gift purchases via partnerships with major platforms. It also introduces personalized tools such as AI-generated podcasts and product discount searches. Compatible with most websites, the feature expands to ride-hailing soon. This positions Copilot as one of the first public AI services for automated task execution, competing with Amazon's Nova Act and others still in development. The rollout begins

Microsoft plans to integrate Copilot Vision into Windows and mobile platforms, expanding its capabilities beyond the Edge browser. The tool now analyzes live video streams on mobile devices, offering practical solutions like plant care advice. Available on iOS and Android via the Copilot app, similar Windows support is coming soon. The Windows version will enable AI guidance for software like Photoshop and photo/webpage analysis. Unlike Recall, Copilot Vision operates without pre-activation. Tes

Microsoft's AI model, Muse, initially part of its research projects, now lets Copilot users play an AI-generated *Quake II* remake in browsers. Part of the "Copilot for Gaming" initiative, this demo showcases basic level designs and improved resolution (640x360). Muse aims to assist developers with prototyping and enhancing classic games for modern platforms. Future plans include releasing more AI gaming experiences via Copilot Labs and introducing a game coaching feature for real-time strategy

Meta launched Llama 4 AI models, powering its AI assistants on platforms like WhatsApp and Instagram. The series includes Llama 4 Scout, a compact model for single GPUs, and Maverick, matching performance with GPT-4o and Gemini 2.0 Flash. Behemoth, still in training, has 288 billion active parameters and aims to surpass GPT-4.5. Meta uses a "Mixture of Experts" architecture for efficiency and will discuss future plans at LlamaCon. Despite being labeled "open source," licensing restrictions limit

Tencent's HunYuan multimodal model update on Yuanbao platform enables processing up to 10 images simultaneously, enhancing multi-format comprehension and reasoning capabilities. The feature is available across mobile, desktop, and web versions, improving efficiency and user experience in content structuring and material generation tasks.

A new study reveals OpenAI's GPT-4.5 excelled in a three-party Turing Test, being perceived as human 73% of the time when adopting specific personas. Participants struggled to distinguish between GPT-4.5 and humans, with the model sometimes identified as human more often than actual humans. The research highlights large language models' ability to mimic human behavior convincingly, though this does not imply true human-level intelligence. The Turing Test reflects evolving perceptions of AI, with

OpenAI secures $400 billion in funding, led by SoftBank, reaching a $3 trillion valuation. The funds will expand computing infrastructure and support the Stargate initiative for AI data centers. With 500 million weekly ChatGPT users, OpenAI aims to advance AI research and tools. Transitioning to a for-profit model is contingent on approvals and overcoming potential legal challenges.

OpenAI's o3 AI model, initially praised for its reasoning capabilities, now faces revised performance estimates showing significantly higher costs. The Arc Prize Foundation updated the cost from $3,000 to potentially $30,000 per ARC-AGI task for o3 high. This highlights the substantial computational resources required, with o3 high using 172 times more resources than o3 low. OpenAI has not finalized pricing, keeping o3 in "preview" status. Rumors suggest expensive enterprise plans, but concerns

Anthropic launched an educational version of Claude, competing with OpenAI's ChatGPT Edu. Tailored for higher education, it includes a Learning Mode to enhance critical thinking and features like enterprise-grade security. Partnerships with Instructure and Internet2, along with university-wide agreements, aim to integrate Claude into educational systems. Despite mixed research on AI's educational impact, Anthropic targets growth in the academic market.

Adobe Premiere Pro 25.2 introduces AI-powered features like Generative Extend, which can add up to two seconds to video clips and generate 4K-quality content. The update also includes an AI-driven search panel, automatic color management, and subtitle translation in 27 languages. After Effects 25.2 adds improved HDR monitoring, animation controls, and optimized preview playback for smoother performance on modern devices.

Google executive Sissie Hsaio is stepping down as head of the Gemini app, with VP Josh Woodward taking over her role. This change aims to focus on the next phase of Gemini's development. Hsaio, a near 20-year Google veteran, will take a break before returning in a new role. Woodward will also continue leading Google Labs and shaping Gemini's future.

Qualcomm acquires MovianAI from VinAI to bolster its AI capabilities in mobile, automotive, and IoT sectors. This move follows prior collaboration on AI solutions like GuardPro. Qualcomm aims to compete with Nvidia by enhancing chip performance and efficiency, introducing the Snapdragon X chipset for mid-range Windows PCs with advanced AI features. Despite strong 2024 financials, Qualcomm faces challenges in the growing AI chip market dominated by Nvidia.

OpenAI announces the upcoming release of a new open language model, inviting global developers to provide feedback. This marks a significant step in the LLM domain, offering users more autonomy and flexibility to modify the model according to their needs. OpenAI will host events in San Francisco, Europe, and the Asia-Pacific region to gather insights and showcase early prototypes. Before launch, the model will undergo evaluation for stability and usability, addressing potential challenges. This

OpenAI's CEO Sam Altman warns of potential product delays due to the overwhelming popularity of ChatGPT and its new image generation tool. Altman notes capacity challenges, despite employees working overtime. OpenAI postponed the free rollout of the image tool and disabled Sora's video feature for new users. ChatGPT added one million users in an hour, reaching 500 million weekly active users and 20 million paid subscribers.

Meta's VP of AI Research, Joelle Pineau, announced her departure after leading FAIR for over two years. Her exit coincides with Meta's $65 billion AI investment plan for 2025. The company is currently searching for a successor as Pineau plans to take a break before starting a new venture.