- NATURAL 20
- Posts
- OpenAI’s Orion Model Revealed
OpenAI’s Orion Model Revealed
PLUS: OpenAI, Anthropic Sign Government Deals, Google's GameNGen Changes Game Development and more.
Use AI as Your Personal Assistant
Ready to embrace a new era of task delegation?
HubSpot’s highly anticipated AI Task Delegation Playbook is your key to supercharging your productivity and saving precious time.
Learn how to integrate AI technology into your processes, allowing you to optimize resource allocation and maximize output with precision and ease.
Today:
OpenAI’s Orion Model Revealed
OpenAI, Anthropic Sign Government Deals
OpenAI Eyes $100 Billion Valuation
Google Unveils Gems, Imagen 3
Google's GameNGen Changes Game Development
Midjourney Expands with Hardware Focus
Artifacts Now on Claude.ai Apps
BREAKING: OpenAI's SHOCKING "ORION" Model! 🔥 Feds get involved 🔥 All details exposed 🔥 It is over…
OpenAI's secretive AI model, formerly called Q STAR and now known as "Strawberry," was recently demonstrated to U.S. National Security officials, confirming its existence. Strawberry's unique technology aims to enable AI to autonomously navigate the internet and perform deep research, not just generate answers.
This model is a precursor to OpenAI's next-generation AI, "Orion," which is being developed to enhance AI reasoning and security. Orion's development may have significant implications for national security, AI safety, and open-source AI.
OpenAI, Anthropic sign deals with US govt for AI research and testing
OpenAI and Anthropic have signed agreements with the U.S. government to test and evaluate their artificial intelligence models. These deals, the first of their kind, aim to ensure safe and ethical AI use amid growing regulatory scrutiny.
The U.S. AI Safety Institute, part of the National Institute of Standards and Technology, will access and test the companies' new AI models before public release, collaborating on research to assess their capabilities and risks.
OpenAI in Talks for Funding Round Valuing It Above $100 Billion
OpenAI is in talks to raise billions in a new funding round, which could value the company over $100 billion. Led by Thrive Capital, which plans to invest about $1 billion, this funding round marks the largest investment in OpenAI since Microsoft’s $10 billion infusion in January 2023.
As competition in the AI sector heats up, with companies like Google, Amazon, and Meta making significant investments in their AI technologies, OpenAI aims to strengthen its position in this rapidly evolving industry, which is seen as key to the future of economic transformation.
Google rolling out Gems and Imagen 3, with people generation, to Gemini Advanced
Google is rolling out new features for its AI model, Gemini, including “Gems” and “Imagen 3.” Gems allow users to create custom versions of Gemini tailored to specific tasks, like writing, coding, or career planning. These tools can remember detailed instructions, help break down complex topics, and offer constructive feedback.
Meanwhile, Imagen 3, a new image generation model, will be available for creating realistic and artistic images, with initial access to Advanced subscribers. Google has improved safety and user experience, ensuring no photorealistic identifiable individuals or inappropriate content are generated. These features will be gradually expanded to more users.
Google’s GameNGen: AI breaks new ground by simulating Doom without a game engine
Google has developed GameNGen, an AI-driven system that simulates the classic game "Doom" without a traditional game engine. GameNGen uses a neural network to generate real-time gameplay at 20 frames per second, relying on an AI diffusion model rather than manually coded software.
This breakthrough could transform the $200 billion gaming industry by reducing development time and costs, allowing smaller creators to produce complex games. Beyond gaming, GameNGen's capabilities could impact industries like virtual reality, autonomous vehicles, and smart cities by enabling realistic, real-time simulations.
Midjourney says it’s ‘getting into hardware’
Midjourney, an AI image-generation platform with over $200 million in revenue and no venture capital funding, is venturing into hardware. The company announced a new hardware team based in San Francisco, likely focusing on projects that could align with its ongoing AI efforts in video and 3D generation.
The move follows the hiring of Ahmad Abbas, an engineer involved in the development of Apple's Vision Pro headset. CEO David Holz, who co-founded Leap Motion, brings additional hardware experience. Despite legal challenges related to its AI training methods, Midjourney is continuing to advance its AI capabilities and expand into new areas.
Artifacts are now generally available
Claude.ai has made its "Artifacts" feature available to all users across its Free, Pro, and Team plans, now accessible on iOS and Android apps. Artifacts provide a creative space where users can instantly see, iterate, and build on their work, turning conversations with Claude into collaborative experiences.
This feature supports various use cases: developers can create architecture diagrams, product managers can design interactive prototypes, designers can build visualizations, and marketers can develop campaign dashboards. Free and Pro users can publish and remix Artifacts within the community, while Team users can share them securely in projects for team collaboration.
🧠RESEARCH
The paper introduces GameNGen, a groundbreaking game engine driven by a neural model, allowing real-time interaction with the video game DOOM at over 20 frames per second on a single TPU. It uses a two-phase training approach involving reinforcement learning and a diffusion model, achieving high-quality, stable generation of game frames that are nearly indistinguishable from real game clips by human raters.
The paper improves SwiftBrush, a text-to-image model, to perform better than multi-step models like Stable Diffusion. By tweaking training methods and adding new techniques, the team achieved top-quality image results, surpassing previous models with a new record in image generation.
The paper introduces BaichuanSEED, a large language model trained using an open-sourced data processing pipeline to address the secrecy around data preparation in AI. This model, built with broad data collection and quality-focused reweighting, achieves strong results comparable to major commercial models without specialized task optimization, showing promise for further enhancements in tasks like math and coding.
The paper introduces "Writing in the Margins" (WiM), a new method for Large Language Models to handle long texts in retrieval tasks more efficiently. This technique involves breaking down texts into segments and using a key-value cache for better processing. WiM improves performance without additional training, showing significant gains in accuracy and F1-score for reasoning and aggregation tasks.
This paper serves as a tutorial for constructing vision-language models (VLMs), which process both image and text inputs to generate text outputs. It begins by reviewing existing approaches, noting their strengths and challenges, and suggesting areas for further research. They also introduce Docmatix, a significantly larger dataset aimed at enhancing document understanding, which accompanies the release of the new model and training datasets.
🛠️TOP TOOLS
Nvidia Nim APIs - Experience the leading models to build enterprise generative AI apps now
Ideate.xyz - AI design co-pilot for content creators & teams to produce scroll-stopping thumbnails for social media channels
Viggle - Powered by JST-1, the first video-3d foundation model with actual physics understanding, starting from making any character move as you want.
Openart AI - Explore, create, and iterate with our intuitive AI drawing tools and editing suite, designed to transform your artistic concepts into reality.
Cape Privacy - Agentic workflow automation across all of your data
📲SOCIAL MEDIA
Today, we are rolling out three experimental models:
- A new smaller variant, Gemini 1.5 Flash-8B
- A stronger Gemini 1.5 Pro model (better on coding & complex prompts)
- A significantly improved Gemini 1.5 Flash modelTry them on , details in 🧵
— Logan Kilpatrick (@OfficialLoganK)
5:09 PM • Aug 27, 2024
🗞️MORE NEWS
Arizona State University personalizes learning and advances research with ChatGPT
Arizona State University (ASU) integrates ChatGPT to enhance education, research, and operations, personalizing learning and fostering innovation. With over 200 projects, ASU uses AI to support teaching, improve outcomes, and ensure ethical practices. ASU’s approach reflects its commitment to inclusion, responsible AI use, and preparing students for future careers. OPENAI
Nvidia launches NIM Agent Blueprints, allowing developers to quickly build enterprise AI apps
Nvidia launched NIM Agent Blueprints, a set of AI workflows and tools to help developers quickly create AI-powered applications for enterprise use. These blueprints cover use cases like customer service avatars, drug discovery, and data extraction. Nvidia plans monthly additions and partners with major firms to simplify access and deployment. VENTUREBEAT
Amazon aims to launch delayed AI Alexa subscription in October
Amazon plans to launch an AI-enhanced version of Alexa in October, featuring daily news summaries, conversational shopping tools, and a child-friendly chatbot. The new version will require a subscription, priced around $10 per month, while the original remains free. This revamp aims to improve engagement and boost e-commerce sales. THE WASHINGTON POST
OpenAI, Adobe and Microsoft support California bill requiring watermarks on AI content
OpenAI, Adobe, and Microsoft back California’s AB 3211 bill requiring AI-generated content to be labeled with watermarks. The bill mandates clear labeling for AI-produced photos, videos, and audio, ensuring transparency for viewers. Initially opposed, the companies now support the revised bill, which aims to prevent misinformation and enhance content authenticity. TECHCRUNCH
Alibaba releases new AI model Qwen2-VL that can analyze videos more than 20 minutes long
Alibaba Cloud released Qwen2-VL, a new AI model capable of analyzing videos over 20 minutes long, understanding multiple languages, and performing real-time video analysis for tasks like tech support. It comes in three sizes, with two open-source variants. Qwen2-VL aims to enhance visual comprehension and multilingual text-image processing across various applications. VENTUREBEAT
With 10x growth since 2023, Llama is the leading engine of AI innovation
Llama, Meta's open-source AI model, has reached 350 million downloads, growing 10x since 2023. Llama 3.1 expanded its capabilities to support 128K context length and eight languages. Partnered with major tech firms, it supports diverse applications, from customer care to AR, demonstrating robust growth and wide industry adoption. META
What'd you think of today's edition? |
Learn AI with us. Let’s Build the Future Together. |
Hello fellow AI-obsessed traveler, Over the past 2 years, as we’ve grown to over 250,000 subscribers between the YouTube Channel and this newsletter, we've received an overwhelming number of requests for one specific thing. While the newsletter helps keep you up to speed with AI news, many of you have asked for the next step: to learn how to actually apply AI in your work. Today we’re finally announcing the solution with NATURAL 20, the community for like-minded AI learners. As a loyal newsletter reader you are getting access at the lowest price it will ever be: JOIN NATURAL 20 AI UNIVERSITY TODAY What you get: * Tutorials by experts across various AI fields. * Daily tutorials by Wes Roth about the latest use cases. * Building Autonomous AI Agents to Automate Your Life and Business (NEW!) * A network of the top 1% of early AI adopters. * Access to community-only resources and software. * And many more features rolling out soon. |
Reply