Microsoft adds AI-powered summarization to Notepad in the Windows Insider program, allowing users to generate customizable summaries via right-click or Ctrl+M. The feature requires a Microsoft account but can be disabled in settings. Additionally, Notepad is testing a recently closed files viewer, while Snipping Tool gains an auto-straightening "draw & hold" feature for screenshots.

Google's Gemini chatbot gains a personalization feature using users' search history to improve responses, powered by the Gemini 2.0 Flash Thinking Experimental model. It offers tailored recommendations for restaurants or travel and may integrate with YouTube and Google Photos. Users can opt-out and view how responses were generated. The feature is rolling out to web and mobile users in over 40 languages. Additional updates include custom AI assistant creation and enhanced integration with Calend

The native image output feature in Gemini 2.0 Flash is now available globally via Google AI Studio. This multimodal tool integrates text and images, supports conversational editing, understands world knowledge, and excels in text rendering. Developers can access it through the Gemini API for use cases like story illustration, ad creation, and interactive applications. Feedback will refine the feature toward a production-ready state.

Open-Sora 2.0, an open-source video generation model with 11 billion parameters, achieves commercial-grade performance at a fraction of the cost. Trained on 224 GPUs for $200,000, it matches models with 30 billion parameters. The release includes weights, inference code, and full distributed training processes, featuring optimizations like low-resolution training and compressed autoencoders to reduce costs and improve speed. This breakthrough lowers barriers for high-quality video generation and

Nvidia's GTC conference begins in San Jose, focusing on AI and accelerated computing. Expect key announcements: the Blackwell B300 series (codenamed Blackwell Ultra) with 288GB memory, previews of the Rubin lineup for 2026, and discussions on post-Rubin products. Quantum computing insights will be shared during "Quantum Day." Despite overheating issues in early Blackwell cards and market competition concerns, Nvidia remains dominant with 82% GPU market share and record revenues. CEO Jensen Huang

Google launched Gemma 3, an updated AI model based on Gemini technology. It supports over 35 languages and excels in text, image, and video analysis, outperforming rivals like Llama and OpenAI on single GPUs. Optimized for Nvidia hardware, Gemma 3 features a upgraded visual encoder for high-resolution images and includes ShieldGemma 2 for safety filtering. Despite concerns about potential misuse of its STEM capabilities, risks are deemed low. The model remains under restrictive licensing but off

Apple added a disclaimer to iPhone 16 series pages about delayed Siri features, stating they will arrive in future updates. This follows acknowledgment of a slower-than-expected AI upgrade process. Apple also removed an ad for the unreleased upgrades, highlighting transparency and commitment to customer experience.

SiliconCloud launches Batch Inference for DeepSeek-R1 and V3API, enhancing large-scale data processing with completion within 24 hours. Batch inference costs 50% less than real-time, with a limited-time 75% discount on DeepSeek-R1. The update includes increased TPM limits and supports diverse functionalities like JSON Mode and FIM, offering efficient, flexible solutions for non-real-time tasks.

GroupMe integrates Microsoft's Copilot feature, enabling users to interact privately with Copilot by long-pressing a message. This update enhances communication efficiency and personalization, allowing users to draft replies or create playlists based on discussions. Copilot expands GroupMe's functionality for a smarter user experience.

Google DeepMind unveiled two AI models, Gemini Robotics and Gemini Robotics-ER, enabling robots to perform diverse real-world tasks. Built on Gemini 2.0, these models advance generality, interactivity, and dexterity, allowing robots to handle novel situations and complex actions like origami folding. Gemini Robotics-ER integrates vision-language capabilities with physical reasoning for tasks such as packing lunchboxes. The company emphasizes safety through a hierarchical approach and new benchma

Snapchat's Platinum subscribers will access new AI-generated video filters with animal and flower themes. Options include a fox on the shoulder, a raccoon on the head, and blooming flowers in front of the user. Powered by Snap's generative video model, these filters will update weekly, advancing Snapchat's use of AR, ML, and AI technologies.

Beijing's Humanoid Robotics Innovation Center launched the "HuiSiKaiWu" universal embodied intelligence platform. It overcomes traditional limitations by moving beyond single-scenario development, filling a gap in general-purpose software for embodied intelligence. This advancement enables robots to autonomously decide and operate in complex environments, marking a key milestone in intelligent robotics technology.

DouBao's large model team released the technical report for Seedream 2.0, an advanced image generation model. It serves hundreds of millions of users, excelling in text rendering, Chinese culture understanding, and bilingual support. Key innovations include a knowledge integration framework, LLM-based bilingual alignment, dual-modal encoding fusion, and DiT architecture upgrades. Post-training RLHF optimization enhances performance through specialized reward models. Evaluated on Bench-240, Seedr

Google owns 14% of Anthropic's shares and plans to invest an additional $750 million, bringing total funding to over $3 billion. Although Google lacks direct control, its backing raises questions about Anthropic's independence amid increasing scrutiny from regulators. Amazon is also investing up to $8 billion, highlighting concerns over AI startups' ties to tech giants.

Manus platform partners with Alibaba's Tongyi Lab to deploy its features using Qwen series open-source models on domestic platforms. The collaboration has begun, with Manus, Monica's AI Agent product, gaining attention for leveraging fine-tuned Qwen models. Alibaba's team looks forward to future partnerships with global AI innovators.

Dexterity, a startup creating industrial robots with human-like flexibility, raised $95 million, reaching a valuation of $1.65 billion. Backed by Lightspeed and Sumitomo, the company focuses on AI-powered robots for warehouse tasks. Meta and Apple are also exploring investments in humanoid robotics as demand grows.

Flower Labs, backed by Y Combinator, launched Flower Intelligence, a distributed cloud platform for AI models. Mozilla uses it for a Thunderbird plugin, leveraging on-device AI with cloud support when needed. The platform focuses on privacy and speed, using local models and switching to Flower's cloud for extra power. It supports open models like Meta’s Llama and includes encryption for data protection. Developers can apply for early access, with more features coming soon. A London summit is pla

OpenAI's CEO Sam Altman revealed a new model excelling in creative writing, capable of producing metafiction stories about AI and sadness. This marks OpenAI's first significant step into narrative writing, signaling potential advancements in literary composition by AI, though release details are unclear.

IBM CEO Arvind Krishna views AI as a valuable tool enhancing productivity rather than replacing jobs, challenging claims that 90% of code will soon be AI-generated. He predicts AI will boost output by 20-30%, emphasizing its role in improving efficiency and quality. While acknowledging concerns about intellectual property and energy consumption, Krishna believes advancements will make AI more affordable and efficient. He distinguishes AI's limitations from quantum computing's potential for break

Meta is testing self-developed chips for AI system training to reduce reliance on external suppliers like Nvidia. These chips, manufactured by TSMC, are designed for specific AI tasks. Previous custom chips were used only for running models, not training. Meta plans to expand production if small-scale tests succeed. With a $650 billion capital expenditure plan for 2023, mostly on Nvidia GPUs, cost savings from in-house chips could significantly benefit the company.