What is the Stargate Project?

On January 21, 2025, President Donald Trump announced the launch of "Stargate," a joint venture among OpenAI, Oracle, SoftBank, and MGX, aiming to invest up to $500 billion in artificial intelligence (AI) infrastructure across the United States by 2029. The project will commence with an initial investment of $100 billion to construct data centers, starting with ten facilities in Texas, and is expected to create over 100,000 jobs. Key technology partners include Microsoft, NVIDIA, and Arm.

The initiative seeks to enhance AI capabilities, with potential applications in digital health records and customized vaccines, and to strengthen the U.S.'s position in the global AI landscape, particularly against competitors like China. The project has garnered support from industry leaders such as Sam Altman of OpenAI and Larry Ellison of Oracle, who highlighted AI's potential to revolutionize disease detection and treatment. However, figures like Elon Musk have expressed skepticism regarding the project's funding.

Read announcement from OpenAI

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Introducing DeepSeek R1

DeepSeek, a Chinese AI research lab backed by High-Flyer Capital Management, has unveiled its latest reasoning models, DeepSeek-R1 and DeepSeek-R1-Zero, as open-source alternatives to proprietary systems like OpenAI's o1. Distributed under the MIT license, DeepSeek-R1 allows developers to freely use, modify, and commercialize the model. The model and its API, which supports fine-tuning and distillation, are accessible at chat.deepseek.com.

Alongside DeepSeek-R1, the lab released six distilled models ranging from 32 billion to 70 billion parameters, optimized for efficiency and claiming performance levels similar to OpenAI's o1-mini. These models are designed to tackle tasks in mathematics, code generation, and reasoning with competitive accuracy. Leveraging large-scale reinforcement learning in post-training, DeepSeek-R1 achieves high performance with minimal reliance on labeled data. The model scored 79.8% Pass@1 on the AIME 2024 benchmark, slightly surpassing OpenAI's o1-1217. This development follows DeepSeek's recent launch of DeepSeek-V3, touted as the best open-source model.

Download from Hugging Face

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Microsoft releases Copilot Chat for Microsoft 365 customers

Microsoft has taken another significant step in its AI journey by launching Copilot Chat, a service aimed at helping businesses integrate AI into their operations seamlessly. This move aligns with the company's broader strategy to empower organizations to adopt AI technologies for enhanced productivity and decision-making. By combining natural language processing with integration into Microsoft 365, the new service is designed to make AI accessible to businesses of all sizes.

Key Features:

• Natural Language Interaction: Users can create AI agents using languages such as English and Mandarin to perform tasks like market research, drafting strategy documents, and meeting preparations.

• Integration with Microsoft 365: While Copilot Chat is free, advanced features—such as summarizing Teams calls and creating PowerPoint slides—require a $30 monthly Microsoft 365 Copilot subscription.

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

GitHub releases Copilot Workspace to public preview

GitHub has officially removed the waitlist for its Copilot Workspace, making the advanced AI development environment available to all developers. The Copilot Workspace integrates agentic AI features to assist with a wide range of development tasks, such as brainstorming ideas, planning, generating and implementing code, iterating and editing, testing, validating, and collaborating with teams. This environment is designed to optimize productivity, particularly for teams, by providing an integrated and collaborative space. It supports natural language queries and can generate code suggestions based on tasks described in plain language, offering an intuitive way to interact with AI.

Additionally, Copilot Workspace enables a smooth workflow by allowing developers to work directly within the environment to build, run, and test code. It also supports the creation of pull requests and version control, simplifying collaboration. The integration of GitHub Codespaces ensures that developers can execute code within a fully functional cloud-based environment. The workspace allows for seamless iteration on projects, making it possible for developers to generate and refine code in real-time, helping teams innovate faster and more efficiently.

Get started with Copilot Workspace

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Nvidia unveils Project DIGITS

At CES 2025, Nvidia unveiled Project DIGITS, a compact AI supercomputer priced at $3,000, designed to empower AI researchers, data scientists, and students. This desktop system is equipped with Nvidia's GB10 Grace Blackwell Superchip, delivering up to a petaflop of AI computing performance. It features a GPU based on Blackwell architecture, 128GB of unified memory, and up to 4TB of NVMe SSD storage. Project DIGITS enables users to run AI models with up to 200 billion parameters, facilitating the development and refinement of complex AI applications without relying on cloud infrastructure.

Nvidia CEO Jensen Huang emphasized that Project DIGITS bridges the gap between high-performance AI computing and accessibility, bringing supercomputer capabilities to individual desktops. The system runs on Nvidia's Linux-based DGX OS and integrates seamlessly with the company's AI ecosystem, offering a flexible development pipeline. Developers can prototype AI models locally on Project DIGITS and scale them to Nvidia's cloud services or data center environments. This initiative aims to democratize AI development, providing powerful tools to a broader range of users.

Learn more about Project DIGITS

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Microsoft releases Phi-4

Microsoft has unveiled Phi-4, a 14-billion-parameter Small Language Model (SLM) that excels in complex reasoning tasks, particularly in mathematics, while maintaining a compact size. This advancement is attributed to the use of high-quality synthetic datasets, curated organic data, and innovative post-training techniques. Benchmark evaluations indicate that Phi-4 surpasses both comparable and larger models, including Gemini Pro 1.5, in mathematical reasoning and problem-solving.

Phi-4 is accessible through Azure AI Foundry under a Microsoft Research License Agreement and is slated for release on Hugging Face. Microsoft emphasizes responsible AI development, offering robust tools within Azure AI Foundry to help organizations measure, mitigate, and manage AI risks throughout the development lifecycle. These tools include Azure AI Content Safety features such as prompt shields and protected material detection, which can be integrated into applications via a single API. This commitment ensures that innovations like Phi-4 are deployed safely and responsibly.

Read the official announcement

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Gemini 2.0 is now here!

Google has unveiled Gemini 2.0, its latest AI model designed to advance autonomous AI capabilities. Building upon the multimodal focus of Gemini 1.0, this iteration emphasizes the development of AI agents capable of understanding complex tasks, anticipating user needs, and performing multi-step actions with minimal human intervention.

Key Features of Gemini 2.0:

• Deep Research: This feature enables Gemini to autonomously gather and synthesize information from the web, producing comprehensive reports that mimic human research methodologies. Available to Gemini Advanced subscribers, it enhances the depth and quality of information retrieval.

• Integration with AI Overviews: Gemini 2.0 enhances Google's AI Overviews in Search, allowing for more sophisticated handling of complex and multimodal queries. This integration aims to provide users with richer and more accurate search experiences.

• Gemini 2.0 Flash: An experimental version designed for high-speed and efficient performance, Flash 2.0 includes native image generation capabilities, improving both the speed and quality of AI-generated images.

• Agent Capabilities: Gemini 2.0 introduces AI agents designed to handle dynamic workflows, automate complex user requests, and adapt to evolving user needs. These agents are capable of independent reasoning, executing intricate tasks, and working across multiple domains to achieve user-defined goals.

In addition to these features, Google introduced Project Astra, an advanced AI assistant capable of real-time interaction with the physical world through devices like smartphones and augmented reality glasses. Project Mariner was also announced, focusing on AI-driven web browsing to autonomously navigate and extract information. Both projects are in early testing phases, with limited public availability anticipated.

Read the official announcement

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

OpenAI’s Sora is now available!

OpenAI has officially launched Sora, its text-to-video AI model, transitioning it from research preview to a standalone product accessible at Sora.com. Sora enables users to generate realistic videos from text prompts, offering features such as video generation up to 1080p resolution, durations up to 20 seconds, and various aspect ratios including widescreen, vertical, and square. Users can also incorporate their own assets to extend, remix, and blend content, or create entirely new videos from text. The platform includes tools like a storyboard feature for precise frame-by-frame input specification and community-driven Featured and Recent feeds showcasing user creations.

(The clip above is taken from Sora.com)

Sora is available to ChatGPT Plus subscribers at no additional cost, allowing up to 50 video generations per month at 480p resolution, with options for fewer videos at 720p. For enhanced usage, the Pro plan offers increased capacity, higher resolutions, and longer video durations. OpenAI plans to introduce tailored pricing for different user categories early next year. To ensure responsible use, Sora-generated videos include C2PA metadata for transparency, and the platform has implemented safeguards against misuse, such as blocking the creation of child sexual abuse material and sexual deepfakes. Initially, uploads featuring people are limited to a small group of early testers to refine deepfake mitigation strategies. OpenAI aims for Sora to inspire new forms of creativity and storytelling, while actively engaging with society to develop norms and safeguards for its responsible deployment.

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

MIT researchers develop new AI tool to predict future floods

Researchers at the Massachusetts Institute of Technology (MIT) have developed an innovative tool that combines generative artificial intelligence (AI) with physics-based flood modeling to produce realistic satellite images depicting potential future flooding scenarios. This method, termed the "Earth Intelligence Engine," aims to assist communities in visualizing and preparing for impending storms by providing detailed, bird's-eye-view images that highlight areas likely to experience flooding based on the projected strength and path of approaching hurricanes.

In a case study focusing on Houston, Texas, the team applied their technique to simulate the aftermath of a storm comparable to Hurricane Harvey, which devastated the region in 2017. The AI-generated images were then compared to actual satellite photographs taken post-Harvey, as well as to images produced without incorporating the physics-based flood model. The results demonstrated that the physics-reinforced AI method generated more accurate and realistic depictions of flooding, whereas the AI-only approach sometimes produced "hallucinations," indicating flooding in areas where it was not physically feasible. This advancement underscores the potential of integrating AI with physical models to create trustworthy visualizations that can enhance public awareness and readiness for natural disasters.

Read the official announcement

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Introducing NVIDIA Fugatto

NVIDIA has introduced Fugatto, a generative AI model capable of creating and transforming a wide array of audio content based on text and audio prompts. This versatile tool can generate music snippets from textual descriptions, modify existing tracks by adding or removing instruments, and alter vocal attributes such as accent and emotion. Notably, Fugatto can produce entirely new sounds, like making a trumpet bark or a saxophone meow, showcasing its innovative approach to audio synthesis.

Designed to understand and generate sound in a human-like manner, Fugatto supports numerous audio generation and transformation tasks. It offers users fine-grained control over audio attributes, enabling the combination of various instructions during inference. Potential applications include rapid prototyping for music producers, personalized language learning tools, and dynamic audio asset creation for video games. This model represents a significant advancement in AI-driven audio technology, providing a new instrument for creative expression.

Read the official announcement

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Copilot Studio updates from Microsoft Ignite 2024

Microsoft has announced significant enhancements to Copilot Studio at the Ignite 2024 conference, aiming to empower organizations in building more intelligent and efficient AI agents. Key updates include:

• Advanced Knowledge Management: Copilot Studio now offers enhanced tuning capabilities and integration with third-party sources. This allows creators to refine their agents using the latest generative models and Retrieval Augmented Generation (RAG) improvements. Users can curate and fine-tune knowledge bases, ensuring agents provide relevant and accurate responses. Additionally, analytics on knowledge sources are now available to monitor performance.

• Azure AI Integration: Makers can directly utilize Azure AI capabilities within Copilot Studio. This includes accessing custom Azure AI Search indexes for tailored RAG scenarios and leveraging over 1,800 prebuilt or custom Azure AI models, enhancing the versatility and intelligence of AI agents.

• Autonomous Agents Preview: A new feature in preview enables the creation of autonomous agents that operate without continuous human prompting. These agents can be built from scratch or configured from prebuilt templates within Copilot Studio, aiming to save time and increase operational efficiency.

• Multimodal Capabilities: Copilot Studio introduces multimodal functionalities, allowing agents to process and generate content across various formats, including text, images, and potentially other media types. This broadens the scope of tasks agents can handle, making them more adaptable to diverse business needs.

Read the official announcement

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

NVIDIA accelerates Google Quantum AI Processor Design

NVIDIA is collaborating with Google Quantum AI to enhance the design of Google's next-generation quantum processors through advanced simulations. Utilizing NVIDIA's CUDA-Q platform and the Eos supercomputer, Google Quantum AI researchers can simulate the physics of quantum devices, addressing challenges such as "noise" that limit current quantum hardware performance. This approach enables comprehensive simulations of 40-qubit devices, significantly accelerating the design process by reducing simulation times from a week to mere minutes.

The CUDA-Q platform, powered by NVIDIA H100 Tensor Core GPUs, facilitates large-scale, realistic simulations of quantum processors. By leveraging this technology, Google aims to scale up quantum hardware while mitigating noise, a critical step toward developing commercially viable quantum computers. The software supporting these simulations will be publicly available through the CUDA-Q platform, providing quantum hardware engineers with tools to rapidly advance system designs.

What is CUDA-Q?

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Introducing Hunyuan

Tencent recently launched Hunyuan, a large language model (LLM) with 70 billion parameters, designed to outperform models like Meta’s Llama 3.1. Optimized for robust multilingual capabilities, Hunyuan excels in both generative and comprehension tasks, especially in Chinese. Tencent has positioned it as a versatile, enterprise-ready tool for applications in sectors like e-commerce, financial services, and content creation, enhancing its competitive edge in the Asian LLM market.

Hunyuan’s architecture focuses on accuracy and efficiency, reducing biases and improving content relevance. Accompanied by the Hunyuan API, this model allows for seamless integration into various applications, giving developers an advanced AI option for multilingual needs. Tencent’s launch highlights its ambitions in the global AI market, offering Hunyuan as a powerful solution that advances AI-powered language understanding and generation.

Try on Hugging Face

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Introducing LucidSim

MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and the Institute for AI and Fundamental Interactions (IAIFI) have introduced LucidSim, a groundbreaking framework that trains quadruped robots, like robotic dogs, to perform parkour and navigate obstacles using only standard RGB camera inputs. Unlike traditional methods that rely on depth sensors or real-world data, LucidSim utilizes generative AI models to create realistic image sequences from the robot's perspective. This innovative approach enables robots to learn dynamic, complex tasks in simulated environments and transfer those skills directly to real-world scenarios without further tuning.

LucidSim’s success was showcased at the 2024 Conference on Robot Learning (CoRL), where researchers demonstrated a robot dog trained entirely in simulation successfully executing parkour maneuvers in various real-world environments. By combining generative models with physics simulators, LucidSim offers a new dimension of versatility and cost-efficiency in robotics training, setting a precedent for future research and applications.

Interested developers and researchers can access the project on GitHub or view live demonstrations on the project’s website.

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Introducing PortaChrome

MIT researchers have developed PortaChrome, a portable light system that uses ultraviolet (UV) and RGB LEDs to alter the color and texture of everyday objects, enabling dynamic, programmable displays on items like clothing and electronics. By applying photochromic dye—a light-sensitive ink—to an object, PortaChrome can activate colors via UV light and desaturate them with RGB LEDs, transforming items with custom designs or real-time data visualizations. This new system allows users to personalize their environment effortlessly, providing a fast and user-friendly alternative to previous projector-based color-changing methods.

The PortaChrome device can adapt to various surfaces, with a flexible design that can wrap around curved items or lay flat. In demonstrations, researchers used PortaChrome to display health metrics on a shirt during a hike and to customize items like headphones and wrist splints with intricate designs. The system’s lightweight, modular construction—featuring a silicone diffusion layer and interwoven LED textiles—makes it compatible with everyday items. MIT envisions future applications where users could quickly reprogram surfaces, creating interactive, customizable designs on clothing, accessories, or even car covers.

Read the official announcement

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Mistral launches new Moderation API

Mistral AI has introduced a content moderation API to enhance safety in AI applications. This service, integral to their Le Chat platform, enables users to detect undesirable text across nine policy categories, including unqualified advice and personally identifiable information (PII). The API offers two endpoints: one for raw text and another for conversational content, both supporting multiple languages such as Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian, and Spanish.

The moderation classifier is designed to provide effective guardrails by addressing potential model-generated harms. Mistral AI has shared performance metrics, including Area Under the Precision-Recall Curve (AUC PR) across various policies, demonstrating the classifier's efficacy. The company is collaborating with customers to develop scalable, lightweight, and customizable moderation tools and is actively engaging with the research community to advance safety measures in the AI field.

Read the official announcement

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

GitHub Copilot allows developers to use Anthropic, Google, and OpenAI models

GitHub has expanded its AI-powered Copilot offerings by integrating new models from Anthropic, Google, and OpenAI. This move allows developers greater flexibility in choosing language models that suit their needs, introducing Claude 3.5 Sonnet, Gemini 1.5 Pro, and OpenAI’s o1-preview. Each model offers distinct strengths, catering to different requirements within code generation, debugging, and development workflows. GitHub Copilot now supports model selection, enabling users to tailor the AI's behavior and adapt to diverse development environments.

Additionally, this expansion aligns with GitHub’s commitment to giving developers control over their tools, enhancing productivity, security, and customization within the coding process. These integrations represent GitHub’s strategic shift towards a multi-model ecosystem, empowering developers to experiment with varied AI functionalities directly in the Copilot experience.

Read the official announcement

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Introducing GitHub Spark

Announced at GitHub Universe this year, GitHub Spark is an AI-powered no-code tool by GitHub Next for creating and sharing personalized micro-apps, or "sparks," with ease. Designed for non-technical users, Spark allows app creation through natural language input, enabling quick, custom solutions for niche needs without coding. Users refine apps through real-time previews, history tracking, and automated revision options, supported by multiple AI models for fine-tuning.

Key Points:

• No-code micro-app creation: Simplifies app building with natural language.

• Real-time previews and revisions: Facilitates iterative development.

• Automatic deployment on all devices: PWAs make apps universally accessible.

• Customizable and shareable: Permissions and remix options enhance collaboration.

• AI model support: Enables tailored app functionality.

GitHub Spark’s managed environment supports automatic deployment on desktop and mobile via Progressive Web Apps (PWAs). It provides a themable design system and persistent data storage, allowing users to easily customize and share apps with controlled permissions and remixing options. This platform fosters collaborative customization, enabling users to build tools that are both specialized and adaptable.

Learn more

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Perplexity announces Internal Knowledge Search and Spaces

Perplexity AI has launched two key features—Internal Knowledge Search and Spaces—focused on enhancing enterprise productivity.

Internal Knowledge Search allows users to simultaneously search both internal documents and live web sources, streamlining research processes for industries such as finance, sales, and HR. This feature aids in blending past materials with real-time insights, helping teams work faster and more effectively.

Spaces introduces a collaborative environment where teams can store documents and configure AI assistants to meet specific project needs. Users have control over the assistant’s behavior, and enterprise customers can opt out of sharing data with Perplexity’s training models, ensuring privacy. Future integrations with Crunchbase and FactSet will further enrich search capabilities for premium users.

Read the official announcement

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.

Endor Labs unveils evaluation tool for open-source AI models

Endor Labs has introduced a new tool aimed at enhancing the evaluation of open-source AI models. This tool applies a rigorous scoring system, assessing key aspects such as security, popularity, quality, and activity. The primary goal is to help developers make better-informed decisions by offering clear insights into the risks and benefits associated with the packages or models they integrate. This evaluation framework allows organizations to prioritize which dependencies align best with their development and security needs.

The broader ambition of Endor Labs' initiative is to improve software supply chain security by offering not only insights but also actionable data. The tool assists in reducing noise during software composition analysis (SCA) processes, filtering out less relevant alerts to allow teams to focus on critical vulnerabilities. By enhancing visibility and control over dependencies and AI models, Endor Labs aims to foster safer adoption of both OSS components and machine learning frameworks, helping teams optimize their software quality without compromising security or productivity.

Try out the tool

Want to help?

If you liked this issue, help spread the word and share One Minute AI with your peers and community.

Share One Minute AI

You can also share feedback with us, as well as news from the AI world that you’d like to see featured by joining our chat on Substack.