Nvidia reportedly acquired AI startup Gretel, which specializes in synthetic data generation. The deal, worth over $320 million, includes integrating Gretel's 80 employees and technology into Nvidia's generative AI services. Founded in 2019, Gretel raised $67 million from investors. This move strengthens Nvidia amid growing reliance on synthetic data for AI training.

Elon Musk's xAI adds image generation to its API with a model named "grok-2-image-1212." Users can generate up to 10 JPG images per request at $0.07 each. The API currently lacks quality or style adjustments, and user prompts may be optimized by a chat model. xAI is exploring revenue streams as it develops Grok 3, discussing a $10 billion funding round, and has acquired a generative AI video startup while expanding its Memphis data center.

Hugging Face argues for open-source and collaborative AI development as a competitive advantage for the US, contrasting with calls for minimal regulation by companies like OpenAI. Their recommendations to the White House's AI Action Plan emphasize democratizing AI through increased research investment, supporting efficient models for small organizations, and proposing transparent systems for enhanced security. While policy divisions grow, Hugging Face suggests leveraging both open-source and pro

OpenAI launched o1-pro, an upgraded AI model offering superior responses via its developer API. Accessible to developers who spent at least $5 on OpenAI services, it is priced at $150 per million input tokens and $600 per million output tokens. Despite leveraging more computational power for deeper reasoning, feedback from ChatGPT Pro users indicates mixed results, with only marginal improvements in some tasks compared to the standard o1 model.

Razer launched Wyvrn, a developer platform integrating AI tools for game development. Key features include AI QA Copilot, which automates bug detection and reduces testing time by 50%, AI Gamer Copilot offering real-time gameplay advice, an updated Chroma RGB platform, THX Spatial Audio Plus for immersive audio, and Razer Sensa HD Haptics enhancing simulations. Partnerships with Side and SimHub aim to expand tool adoption and capabilities.

Microsoft resolved a bug causing Copilot to uninstall on some Windows 11 devices. Affected systems are being restored, reinstating the AI assistant. Previously, users had to reinstall Copilot manually. This month, Copilot was updated with a cleaner UI and voice chat support. Another issue prevented Roblox downloads on Arm-based devices, which is being addressed by Roblox.

Tencent released five Hunyuan3D-2.0 open-source models, including the Turbo series for accelerated geometry generation, multi-view versions for enhanced detail capture, and a lightweight mini series for reduced computational costs. The upgraded 3D AI creation engine supports multi-view input, improves mesh reduction and PBR material generation for realistic textures, and ensures compatibility with multiple formats like OBJ, GLB, and FBX. These advancements streamline 3D model creation for applic

Google is developing open AI models called TxGemma for drug discovery, set to release later this month. These models can understand regular text and therapeutic structures like chemicals and proteins. While aiming to enhance drug development efficiency, Google has not clarified the commercial use or customization terms of these models. Despite past failures in AI-driven drug discovery and variability in system accuracy, major pharmaceutical companies and investors remain optimistic about the tec

Anthropic is developing voice interaction features for its AI chatbot Claude, aiming to enhance user experience through voice-based AI model interactions. The company explores integrating this technology into desktop versions and has created initial prototypes. Discussions with Amazon and ElevenLabs are underway, though no formal agreements have been reached. Anthropic is collaborating with multiple partners to accelerate the development and market launch of these features.

Meta's Llama AI models hit 1 billion downloads, up 53% since early 2024. Key to Meta's AI strategy, Llama powers platforms like Facebook and Instagram, though licensing restrictions have drawn criticism. Integrated by firms like Spotify and DoorDash, Llama faces challenges including copyright lawsuits and EU privacy concerns. Meta plans to invest $80 billion in AI this year, refine Llama using insights from rivals like DeepSeek, and launch new versions with advanced capabilities. Details will be

Google's Gemini chatbot introduces Canvas, an interactive workspace for writing and coding projects. It allows users to create, refine, and share content with precision and real-time previews. Features include tone adjustments, collaboration tools, and exporting to Google Docs. Canvas also supports code previews for web apps and integrates NotebookLM's audio summary feature for podcast-style summaries. Both features are free worldwide but have platform and language limitations.

Stability AI launched Stable Virtual Camera, an AI model transforming 2D images into immersive videos with realistic depth. It supports dynamic camera paths and various aspect ratios but may produce lower quality results with complex scenes or specific subjects. Available under a non-commercial license, this research preview follows previous models like Stable Diffusion and new leadership for the company.

GM and Nvidia expand their partnership to integrate AI across operations, including factories, robotics, and autonomous vehicles. Nvidia will provide GM with AI infrastructure, GPUs, and assistance in developing custom AI systems. This collaboration involves using Omniverse for factory digital twins, Cosmos for manufacturing advancements, and Drive AGX for in-vehicle hardware. The focus is on enhancing production, robotic capabilities, and advancing driver-assistance systems toward full autonomy

Nvidia is collaborating with Disney Research and Google DeepMind to develop Newton, a physics engine simulating robot movements in real-world environments. Disney will use Newton for its next-gen entertainment robots like BDX droids, planning a global theme park rollout starting next year. Newton, set for an open-source release in 2025, enhances robot expressiveness and precision in complex tasks, supports deformable object interactions, and integrates with Google DeepMind's MuJoCo. At GTC 2025,

Nvidia unveiled two personal AI supercomputers, DGX Spark and DGX Station, at GTC. DGX Spark, priced at $3,000, is a compact desktop device delivering up to 1,000 TOPS for AI tasks. DGX Station targets developers and researchers with 20 petaflops of AI performance and 784GB unified memory. OEM partners like Asus, Dell, and HP will produce versions of these devices. Nvidia competes with AMD's Ryzen AI Max+ "Strix Halo" in offering GPUs with significant unified memory for local LLMs.

Nvidia unveiled Groot N1 at GTC 2025, a general-purpose AI model for humanoid robots. It features a dual-system architecture for fast and slow thinking, expanding applications beyond industry use. Open-sourced with a simulation framework, Groot N1 aims to advance humanoid robotics despite historical challenges in mass production.

At Nvidia GTC 2025, Jensen Huang unveiled upcoming GPU products. The Vera Rubin GPU, launching in 2026, features dual GPUs with over 50PFLOPS AI inference performance and a self-designed Vera CPU. Blackwell Ultra, releasing in 2025, boosts memory to 288GB while maintaining 20PFLOPS. Rubin Ultra (2027) will integrate four GPUs for 100PFLOPS, and Feynman (2028) succeeds Vera Rubin with an integrated Vera CPU.

The supercomputing internet platform has upgraded its MaaS service by integrating the QwQ-32B model, known for excellence in mathematical reasoning and code generation. Users can now access multi-turn interactions and long-text parsing via a new Chat feature for diverse scenarios like research and consultations. Additionally, the platform's model API has surpassed a million calls, with new users receiving a free quota of one million tokens to encourage exploration of complex reasoning applicatio

Mistral AI releases open-source Mistral Small 3.1, a lightweight AI model with 24 billion parameters outperforming similar models from OpenAI and Google. It excels in text and image processing, supports up to 128,000 tokens, and processes data at 150 tokens per second. Optimized for modest hardware, it expands accessibility to advanced AI. Part of a diverse product lineup, Mistral Small 3.1 is available on Huggingface, Mistral's API, and major cloud platforms.

OpenAI is testing ChatGPT Connector, a feature allowing business users to integrate Slack and Google Drive with ChatGPT. Subscribers can link their accounts, enabling the chatbot to answer queries based on internal files and conversations. The tool may expand to platforms like Microsoft SharePoint and Box. Powered by GPT-4o, it optimizes responses using company data while respecting permissions. Images and detailed sheet analysis aren't supported, and private Slack messages are inaccessible. Com