Google introduced Lens Screen Search for iPhone users via the Google app and Chrome browser, allowing quick searches by selecting screen content with gestures. This feature resembles Android's "circle search" but is limited to Google apps on iOS. Users can search text, images, or videos without leaving the app and may see AI-generated summaries in results. Future updates will include a Lens icon in the address bar for easier access.

Microsoft unveiled Muse, an AI model aiding Xbox game development by generating game environments, understanding 3D worlds, and responding to player interactions. Trained on data from "Bleeding Edge," Muse processes visual input but currently outputs low-resolution graphics. It supports early game development, enhances classic games, and aids in prototyping. Xbox CEO Phil Spencer envisions AI enabling old games to run on modern platforms without original hardware. While exploring AI's potential,

Mira Murati, ex-CTO of OpenAI, launches Mind Machines Lab, an AI startup focusing on multimodal models. The team includes former OpenAI leaders and aims to develop customizable, open-source AI technology while prioritizing infrastructure quality and safety. Initially backed by over $100 million in funding discussions, the company plans to advance research through community collaboration and publications. With 29 employees, it targets broad human expertise applications beyond programming tasks.

Meta will host LlamaCon on April 29, 2025, focusing on open-source AI development for developers. The event precedes Meta Connect in September, which highlights Meta Horizon advancements. LlamaCon's name derives from Meta's Llama AI model, with more details to come. Meta plans significant AI investments and may release new smart glasses. Ray-Ban Smart Glasses, launched in 2023, have sold over two million units.

Google's AI assistant Gemini may soon gain video generation capabilities, as indicated by code descriptions found in the Google app's APK file. The term "videogen" and Gemini's internal code name "robin" suggest an upcoming feature allowing users to create videos via simple AI commands. Although still under development, this advancement could revolutionize video creation while raising concerns over copyright and privacy. Google has yet to officially confirm the update.

HP acquires majority of Humane's business for $116 million, including its CosmOS system, tech team, and over 300 patents. Humane will discontinue the AI Pin, with server support ending on February 28, 2025, impacting cloud-dependent features. Refunds are limited to recent purchases or prorated subscriptions. Initially priced at $1 billion, weak sales led to the lower acquisition. Humane's founders will lead HP's new AI division, HP IQ, integrating AI into PCs, printers, and meeting solutions.

Gemini's Deep Research feature is now available on mobile devices, enhancing convenience for Gemini Advanced users. Originally launched for web, it provides comprehensive market analysis and aids investment decisions. Users can access detailed financial insights on Android or iOS anytime, anywhere, with web-based functionality still fully supported.

Google Meet enhances its Gemini-powered note-taking tool with a new action item feature, generating next steps, assigning tasks, and setting deadlines post-meeting. Building on last year's voice-to-text functionality, this update aims to improve team collaboration efficiency through AI-driven automation. The rollout begins today but will proceed slower than usual to ensure quality.

Musk unveiled Grok 3, an AI model with a training cost equivalent to 200,000 NVIDIA GPUs. Trained at xAI's advanced Colossus data center, Grok 3's scale is ten times larger than Grok 2. This significant investment in resources could lead to major advancements in reasoning, comprehension, and content generation, potentially reshaping the AI landscape.

Dark Moon Corporation launched Kimi-Latest, a new visual model in the Kimi assistant line. It features advanced context processing up to 128k tokens, automatic model selection for billing, image understanding, and cost-saving caching mechanisms. Retaining functionalities from moonshot-v1, it suits chat apps and AI assistants but may require prompt revisions. For intent recognition tasks, moonshot-v1 is preferable. Note: Kimi-Latest only supports standard Kimi models; API access to Kimi k1.5 requ

Grok 3 developed a 3D model visualizing Earth-to-Mars space travel, featuring accurate planetary representations and a simulated spacecraft journey. The model also demonstrates the return trip, showcasing its reliability. SpaceX plans to send Optimus and Grok to Mars via Starship within two years, advancing robotics in space exploration.

Elon Musk's xAI launched Grok 3, an advanced family of large language models surpassing competitors in math, science, and code writing. Key features include DeepSearch for summarized web queries, optimized reasoning variants, and a subscription service called SuperGrok. Developed using Colossus supercomputers with 100,000 NVIDIA H100 GPUs, Grok 3 offers faster performance and improved accuracy. Voice mode is delayed due to unresolved issues, while Grok 2 will be open-sourced after Grok 3 stabili

Stairway Stars and Geely Automotive Group announced the global open-source release of their co-developed Step series multimodal models. The series includes Step-Video-T2V, a 30-billion-parameter video generation model, and Step-Audio, an industry-first open-source voice interaction model. Step-Video-T2V generates high-quality 540P videos, while Step-Audio supports emotional, personalized voice interactions with natural expression. Users can explore Step-Audio via the Yuewen App, advancing AI inn

Kunlun Tech introduced two AI video technologies: SkyReels-V1, an open-source model for realistic short drama creation with detailed human performances, and SkyReels-A1, an algorithm enhancing facial and motion control. Both aim to reduce costs, increase accessibility, and improve efficiency in AI-driven video generation, enabling broader applications in content creation.

Researchers from the University of St. Louis developed DarkMind, a stealthy backdoor attack targeting large language models (LLMs). It manipulates the reasoning process without detection, activating during specific steps to alter outputs. Unlike traditional attacks, DarkMind embeds hidden triggers within custom LLMs, remaining dormant until triggered. This vulnerability affects advanced models like GPT-4o and LLaMA-3, posing risks to security and reliability across various domains.

Paris-based AI startup Mistral unveiled Mistral Saba, a specialized model for Arabic-speaking countries with 2.4 billion parameters. It outperforms Mistral's general-purpose models in Arabic and Indo-Aryan languages. Saba aims to enhance conversational support and content generation while targeting Middle Eastern clients. Accessible via API or internal deployment, it caters to sectors like finance and healthcare. This launch expands Mistral's multilingual focus and strategic presence in the Midd

Apple's Image Playground app faces criticism for racial bias, struggling to recognize a machine learning expert's skin tone and hair texture. Despite limiting functionality to cartoon-style images, biases persist. This highlights ongoing concerns about fairness and accuracy in AI image generation as a critical issue requiring attention.

Elon Musk's xAI to unveil Grok 3, billed as the "smartest AI on Earth," featuring advanced reasoning and multimodal capabilities. Integrated into X platform, Grok 3 follows Grok 2, which improved real-world performance. Competing with OpenAI's GPT 4o upgrades, xAI negotiates $5 billion in server deals with Dell and raises $10 billion for GPU investments, though its $51 billion valuation lags behind OpenAI's $150 billion.

Tencent has integrated the DeepSeek-R1 model into multiple products, including Tencent Yuanbao, WeChat, ima, Tencent Docs, QQ Browser, and QQ Music. This enhances services with advanced AI capabilities like intelligent search, creative assistance, and deep-thinking problem-solving, showcasing Tencent's AI innovation and commitment to improving user experiences.

Baidu Search announces an update integrating DeepSeek and Wenxin Large Model for enhanced search experiences. These features, available to all users, offer accurate information retrieval and support multi-modal input/output. Developers on the Wenxin platform can leverage DeepSeek to optimize products, boosting ecosystem growth. This integration promises improved efficiency and innovation for both users and developers.