Good morning, it’s Tuesday! Today’s lineup is packed. OpenAI’s Sam Altman is aiming to supercharge global AI with a bold vision for chip factories and data centers backed by hundreds of billions of dollars. Google just shelled out $2.7 billion to reel AI pioneer Noam Shazeer back in, and NVIDIA’s on another AI shopping spree. Meanwhile, California’s governor gives the thumbs down to an AI safety bill, and Hugging Face hits a milestone with one million models. Let’s dive in!

Your Daily Roundup:

1. OpenAI's Bold Plan to Revolutionize Global AI Infrastructure

2. Google Rehires AI Pioneer Noam Shazeer in $2.7 Billion Deal

3. California Governor Vetoes AI Safety Bill Targeting Large Models

4. NVIDIA Acquires OctoAI to Dominate Enterprise Generative AI Solutions

5. Hugging Face Surpasses One Million AI Models on Open-Source Platform

👉️ Top AI Stories

OpenAI's Ambitious Plan to Revolutionize AI Infrastructure

Sam Altman, CEO of OpenAI, has proposed a bold plan to build global AI infrastructure resembling the spread of electricity, envisioning new chip factories and data centers powered by trillions of investment dollars. Despite initial resistance and scaling down his ambitions to hundreds of billions, Altman continues efforts to unite tech companies, investors, and governments to boost computing power critical for advancing AI. → Continue reading here.

Google Rehires AI Pioneer Noam Shazeer for $2.7 Billion Amid AI Talent Race

Google has rehired Noam Shazeer, a key figure in AI who co-authored a foundational research paper on transformers, for $2.7 billion after he left the company in 2021. Shazeer had quit in frustration when Google declined to release a chatbot he developed, but when his startup, Character.AI, struggled, Google made an expensive move to bring him back, reflecting the intense competition for top AI talent. → Continue reading here.

California Governor Vetoes AI Safety Bill Targeting Large-Scale Models

Governor Gavin Newsom vetoed California's AI safety bill (SB 1047), citing concerns that it only targets large, expensive generative AI models while leaving smaller models unregulated. The decision reflects the complexity of establishing comprehensive regulations for AI across the diverse landscape of model sizes and capabilities. → Continue reading here.

NVIDIA Acquires OctoAI to Dominate Enterprise Generative AI Solutions

NVIDIA has reinforced its leading position in the AI industry by acquiring OctoAI, a company that specializes in generative AI tools, for $250 million, marking its fifth acquisition in 2024. The deal enhances NVIDIA's enterprise AI offerings, combining OctoAI's hardware-agnostic model optimization technology with NVIDIA's AI infrastructure, expanding NVIDIA's market reach and addressing the full AI lifecycle from development to deployment across various hardware platforms. → Continue reading here.

Hugging Face Reaches Milestone of One Million AI Models on Its Open-Source Platform

Hugging Face, an AI hosting platform, surpassed one million downloadable AI models, highlighting the rapid growth and interest in machine learning. The company, which shifted from a chatbot app to an open-source hub in 2020, offers specialized models for diverse use cases and continues to expand as a community-driven platform for AI researchers and developers, with a new repository created every 10 seconds. → Continue reading here.

☝️ Sponsor: VULTR

Vultr is empowering the next generation of generative AI startups with access to the latest NVIDIA GPUs. Try it yourself when you click here and use promo code "BERMAN300" for $300 off your first 30 days.

👾 Forward Future Original

No, not life! We haven’t gone far enough to answer that question! Right now, the more pertinent question here, and one that’s easier to answer, is: how does AI, specifically how do large language models (LLMs), represent meaning? 

In my previous article, we touched upon the concept of a bit: a binary unit of information, that simply encodes a true/false, yes/no, I do/I don’t kind of message. But, as we know, there is much more to reality than just ‘I do’! 

The next step from a simple 0 or 1 expression is, of course, bigger numbers, the basic arithmetic that we’re all taught fairly early in our childhood. This in itself allows us to answer a simple question such as ‘How many apples are there in that bag?’. Or a dandier one such as ‘What is the torque on the third nut on the engine of the Falcon 9 rocket at a given altitude?’, which can be handy if you’re going to be interviewed by Elon Musk! 

To help visualize numbers, we can plot them on a number line. Here of course we are enlisting the services of geometry, which is very helpful in developing a better understanding of this and the other concepts we will touch upon here. 

→ Continue reading here. 

🚀 Launches + Funding

• Airtable Launches AI for Enterprise Workflows: Airtable's new AI platform helps businesses build custom AI apps and manage large datasets to streamline operations.

• Microsoft Relaunching AI Screenshot Tool ‘Recall’: Microsoft will relaunch 'Recall' in November 2024 with improved privacy controls and encryption, though concerns about data sharing persist.

• Workera Launches Sage, an AI Mentor: Workera's AI mentor Sage provides personalized skill assessments, feedback, and strategic insights for both individual and enterprise development.

• Harmonic Raises $75M To Advance AI-Driven MSI: Harmonic secures $75M to develop Mathematical Superintelligence (MSI), which enhances AI reliability with formal verification and synthetic data for use in engineering, design, and medical fields.

✌️ Sponsor: Langtrace AI

Monitor, Evaluate & Improve Your LLM Apps

Open source LLM application observability, built on OpenTelemetry standards for seamless integration with tools like Grafana, Datadog, and more. Now featuring Agentic Tracing, DSPy-Specific Tracing, & Prompt Debugging Modes, Langtrace helps you manage the lifecycle of your LLM powered applications. Delivering detailed insights into AI agent workflows, helping you evaluate LLM outputs, while tracing agentic frameworks with precision. Star Langtrace on Github!

✍️ Editor Picks

Research

AI Model Predicts Structures of Crystalline Materials from X-ray Data

MIT researchers have developed an AI model, Crystalyze, that predicts the structures of crystalline materials from powder X-ray diffraction data, helping researchers better understand materials for use in batteries, magnets, and more. The AI model, trained on thousands of materials, generates potential structures from diffraction patterns and has already solved over 100 previously unsolved structures. This tool promises advancements in material sciences, especially in fields reliant on crystalline properties. → Continue reading here.

Papers

Foundation Model for Weather and Climate Breaks New Ground

Prithvi WxC is a 2.3 billion parameter AI foundation model designed to tackle multiple weather and climate prediction tasks, such as forecasting, downscaling, and extreme event estimation. Built on NASA's MERRA-2 dataset and utilizing a transformer-based architecture, Prithvi WxC demonstrates strong performance in zero-shot forecasting and fine-tuning for specific tasks like downscaling and gravity wave flux parameterization, offering significant improvements over traditional numerical models and interpolation baselines. → Continue reading here.

Robotics

NVIDIA's ReMEmbR Enables Robots to Reason and Act Using Generative AI

NVIDIA's ReMEmbR project combines Large Language Models (LLMs), Visual Language Models (VLMs), and retrieval-augmented generation (RAG) to allow robots to reason and act autonomously during long-term deployments. Using NVIDIA's Isaac ROS framework, a Nova Carter robot collects visual data and stores it in a memory database, enabling it to respond to user queries and generate navigation goals—such as finding a snack—by reasoning over its past observations. → Watch it here.

Models

Liquid AI Unveils LFM Models, Surpassing Performance Benchmarks with Efficient Long-Context Processing

Liquid AI introduces its groundbreaking LFM models, setting new standards in AI performance. The LFM-1B and LFM-3B outperform transformer-based models, with LFM-40B offering unmatched efficiency in large-scale processing while maintaining a smaller model size. Liquid AI’s advancements in algorithmic design enhance memory efficiency, long-context capabilities, and adaptability across industries. Their models now challenge larger competitors on key benchmarks, positioning Liquid AI as a leader in scalable AI solutions. → Continue reading here.

📽️ New Videos:

🧰 AI Toolbox

• Ayraa 2.0: AI Tool Boosting Team Productivity Across Apps: Ayraa 2.0 is a generative AI platform that enhances team productivity by organizing and analyzing work data from various apps. Key features include instant search, AI-powered transcriptions, and personalized insights for tasks like customer support and sales.

• BeforeSunset AI: AI-Powered Task Planning: BeforeSunset AI is a productivity tool that uses AI to optimize daily schedules, helping users plan and complete tasks more efficiently. It offers features like AI-powered scheduling, to-do management with reminders, weekly and monthly views, and the ability to automatically move incomplete tasks.

• panda{·}etl: Automating Document Workflows: panda{·}etl is a platform designed to automate document-intensive workflows, enabling users to effortlessly extract, transform, and organize data from PDFs, spreadsheets, and various other file types.

• MIMO: Generalizable Model for Controllable Character Video: MIMO is a cutting-edge model designed for controllable video synthesis, allowing users to create realistic, animatable avatars in complex, interactive scenes. By decomposing video clips into spatial components using monocular depth estimation, MIMO enables flexible control over character identity, motion, and scene interactions.

🛰️ Houston, We Have More Headlines!

• AlphaChip revolutionizes chip design: AlphaChip uses AI to accelerate chip design, achieving superhuman efficiency, advancing AI industry.

• OpenAI's Shift to For-Profit: OpenAI is transitioning to a for-profit corporation, aiming to streamline operations and raise $6.5 billion.

• AI-Generated "Slop" Floods Internet: Low-quality AI-generated content is overwhelming the internet, cluttering search results and platforms.

• Google Expands Data Centers: Google is investing $3.3 billion in Dorchester, SC, to expand its cloud infrastructure with two new data centers.

• AI Defeats CAPTCHA 100% of the Time: Researchers developed an AI, YOLO, capable of solving Google's reCAPTCHAv2 with 100% accuracy.

• ChatGPT Memory Vulnerability: A security flaw in ChatGPT allowed a hacker to plant false memories, potentially stealing user data.

• OpenAI Agrees to Data Inspection: OpenAI will allow inspection of its training data in response to authors' copyright infringement lawsuits.

• China Blocks ChatGPT and AI Tools: China blocked AI platforms in March 2023, likely due to concerns over politically sensitive content.

🥰 Help Us Improve

Reply to this email if you have specific feedback to share. We’d love to hear from you!

🌐 Stay Connected

Looking for more AI news, tips, and insights? Follow us on X for quick daily updates and bite-sized content.

For in-depth technical analysis, subscribe to the Forward Future YouTube channel. We dive deep into new models, test their performance, explore the latest tools, and share our impressions of AI innovations and developments.

Prefer using an RSS feed? Add Forward Future to your feed here: RSS Link

Thanks for reading this week’s newsletter. See you next time!

🧑‍🚀 Forward Future Team