Good morning, it’s Tuesday! In this edition, we have a packed lineup of AI stories, along with the results from last week’s poll: 'What concerns you most about the rapid development of AI?'
Meta’s been feeding its AI all your public posts since 2007, Nvidia's CEO sketches the future of AI infrastructure, and AI's role in stopping skin cancer could change how we think about detection and prevention. Plus, an AI-powered Codeforces contestant stirs the competitive programming pot, while Fei-Fei Li’s new venture might just make AI see in 3D.
Your Daily Roundup:
World Labs Launches with $230M to Pioneer 3D AI for Virtual Worlds
AI Model Achieves Near Master-Level Performance in Codeforces Contest
Meta Used Public Facebook and Instagram Data to Train AI Since 2007
ChatGPT Surpasses 11 Million Subscribers, Generating $2.7 Billion Annually
AI Enhances Skin Cancer Prevention Through Early Detection
Nvidia CEO Jensen Huang Predicts Future AI Infrastructure Transformation
OpenAI o1 models shine in ARC Prize evaluations, sparking discussions on chain-of-thought scaling towards AGI.
👉️ Top AI Stories
World Labs Launches with $230M Funding to Pioneer 3D Perceptive AI
World Labs, co-founded by AI pioneer Fei-Fei Li, has launched with over $230 million in funding and a valuation exceeding $1 billion. The company aims to develop "spatial intelligence" systems called "Large World Models" (LWMs) to enable AI to perceive and interact with the world in 3D, potentially revolutionizing fields such as robotics, virtual worlds, and creative industries by creating immersive virtual environments. Continue reading here.
Codeforces Contestant Uses AI Model to Achieve Near Master-Level Performance
A Codeforces contestant utilized the o1-mini AI model during a live contest, achieving performance close to the master level. The incident has sparked discussions on the use of AI in competitive programming. Moving forward, there is consensus on the need to restrict AI in competitions, and it will be intriguing to observe how the competitive programming landscape adapts to these changes. Continue reading here.
Meta fed its AI on almost everything you’ve posted publicly since 2007
Meta has admitted to using all publicly available text and photos from adult Facebook and Instagram users since 2007 to train its artificial intelligence models, a practice confirmed during a government inquiry in Australia. While the company claims not to collect data from users under 18, questions remain about the handling of data for accounts created when the user was a minor, and only European users currently have the option to opt-out due to local privacy laws. Continue reading here.
ChatGPT Reaches Over 11 Million Paying Subscribers, Generating $2.7 Billion Annually
OpenAI's ChatGPT has reportedly surpassed 10 million paying subscribers, with an additional 1 million users on higher-priced business plans, contributing to an estimated monthly revenue of $225 million. This brings the annual revenue to approximately $2.7 billion, as users leverage the AI for tasks such as generating code, writing, editing, and research. Continue reading here.
AI's Role in Skin Cancer Prevention Through Behavior Change
AI is increasingly aiding in the early detection and prevention of skin cancer by making diagnostic tools more accessible and encouraging proactive healthcare behavior. A 2024 Stanford study showed that AI-assisted clinicians had a higher sensitivity (81.1%) in diagnosing skin cancer compared to those without AI (74.8%). With the development of AI-powered tools like Bupa's Blua, individuals can now assess skin lesions using a smartphone app, significantly increasing the likelihood of early detection and intervention, which is crucial given the rising incidence of skin cancer globally. Continue reading here.
Nvidia CEO Jensen Huang Forecasts the Future of AI Infrastructure
Nvidia CEO Jensen Huang emphasized the importance of AI infrastructure over individual chips at the Goldman Sachs Communacopia + Technology Conference. He highlighted Nvidia's strategic focus on two major trends: transitioning from CPU-based data centers to GPU-accelerated computing for handling large datasets and advancing generative AI, which allows computers to produce original content autonomously. Nvidia's new Blackwell platform, designed to handle trillion-parameter models, is part of this future vision, aimed at creating "supercomputers" within data centers. Continue reading here.
OpenAI's o1 Model: A New Paradigm in Chain-of-Thought Reasoning but Still Limited on ARC-AGI
OpenAI's o1-preview and o1-mini models were tested on the ARC-AGI public evaluation dataset and showed improvements over previous models like GPT-4o, particularly by utilizing an advanced chain-of-thought (CoT) reasoning process. The models excel in tasks requiring structured reasoning but face limitations in generalization on more complex problems, as their accuracy scales with increased test-time compute. While promising in reasoning efficiency, o1 highlights the challenges of developing AGI, pointing to the need for novel approaches beyond current methods. 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 and use promo code "BERMAN300" for $300 off your first 30 days.
👾 Forward Future Original
The Power of Custom Instructions in ChatGPT
ChatGPT has proven to be an incredible asset across nearly every profession and industry. However, those who rely solely on its default settings are missing out on its full potential.
Custom instructions offer a way to transform ChatGPT from a general-purpose language model into a highly specialized tool tailored to your specific needs. Think of them as a blueprint for ChatGPT's output, guiding it to generate text that aligns perfectly with your brand voice, formatting requirements, or even complex multi-step tasks that demand domain expertise.
🚀 Launches + Funding
Luma AI's Dream Machine API Launch: Luma AI's Dream Machine API enables easy integration of video generation capabilities into applications for developers.
Compute.Hyper.Space — Affordable AI Interface: HyperspaceAI offers an affordable AI interface using a reasoning model and open-source tools for customizable AI experiences on consumer devices.
Vchitect-2.0's Text-to-Video Model: Vchitect-2.0 introduces a model converting text and images into videos using a parallel transformer architecture, enhancing video generation efficiency.
Tencent's GameGen-O — Open-World Game Generation: Tencent's GameGen-O is a diffusion transformer model for generating open-world video games, simulating features like characters and environments.
Supermaven Raises $12M Led by OpenAI: Supermaven secured $12 million in funding with contributions from OpenAI and Perplexity co-founders, offering advanced code generation with its Babble model.
✌️Sponsor: Weights & Biases
Weave is a lightweight toolkit for tracking and evaluating LLM applications, built by Weights & Biases. Make real progress on your LLM development, get started with Weave for free today.
✍️ Editor Picks
Politics
AI's Potential Role in Creating Engineered Pathogens Poses Urgent Global Threat
As the capabilities of artificial intelligence (AI) and genetic engineering converge, there is an increasing risk that the next pandemic could be caused by an engineered virus. The ease of modifying viruses for therapeutic purposes now also poses a potential for misuse, as AI enables the creation of novel pathogens beyond human-inspired designs. This risk is amplified by the accessibility of the necessary tools, raising concerns that a state-sponsored program or even an individual could intentionally or accidentally initiate a catastrophic pandemic.
Robotics
Muscle-Powered Robotics: A New Frontier in Biomimetic Engineering
Researchers at ETH Zurich and the Max Planck Institute have developed a robotic leg powered by electro-hydraulic actuators, mimicking biological muscles rather than relying on traditional motors. This breakthrough offers significant improvements in energy efficiency, adaptability, and simplified movement control, allowing robots to react more fluidly to uneven terrains and obstacles. While the technology is still in early stages and not yet fully mobile, it represents a major leap toward creating more lifelike, agile robots with potential applications in rescue operations and human-robot collaboration.
Medical
AI Tool Cuts Unexpected Hospital Deaths by 26%, Canadian Study Shows
Researchers at St. Michael's Hospital in Toronto developed an AI-based early warning system called Chartwatch, which has reduced unexpected deaths by 26%, according to a new study. Chartwatch monitors patients' vital signs and lab results in real-time, predicting potential deterioration and allowing clinicians to intervene earlier. While the technology shows immense potential for improving patient outcomes, its effectiveness still needs to be tested across multiple hospitals before wider implementation.
📽️ New Video:
🧰 AI Toolbox
Algorithm Enhances Emotion Detection Accuracy: Lanbo Xu's research at Northeastern University uses a convolutional neural network and a "chaotic frog leap algorithm" to refine facial features for real-time emotion detection in video, achieving up to 99% accuracy with applications in mental health, security, and marketing.
Nanotech E-Nose for Real-Time Gas Detection: UNIST researchers created an ultra-small e-nose using a nanoheater-based gas sensor and deep learning, operating on less than 200 microwatts to identify gases in real-time, suitable for portable and IoT applications.
👻 Forward Future Poll: The Results
Last week, we asked you what concerns you most with the rapid development of AI. Well, the results are in!
The top concerns are AI-driven misinformation and Autonomous weapons, both at 26%, followed by Loss of control at 22%. While Ethical dilemmas is the least cited concern, with only 4% of respondents.
Here’s the full breakdown:
🛰️ Houston, We Have More Headlines!
g1 Model with Llama-3.1: The model tackles logical problems with improved strategies, achieving 70% accuracy on the Strawberry problem.
Kolmogorov-Arnold Network Launched: Offers greater interpretability while maintaining the performance of traditional neural networks.
Checkbox AI Ethics Reviewed: The paper critiques the implementation of ethical AI in health, highlighting the lack of standardization and guidance.
AMD's AI Frame Generation: AMD's FSR 4 introduces AI-powered frame generation, aiming to improve performance and battery life.
Venado AI Supercomputer Unveiled: Venado combines NVIDIA GH200 Grace Hopper superchips and Grace CPUs, offering advanced capabilities.
Chipotle Tests Food Robots: Chipotle is testing robots for bowl and salad preparation to enhance efficiency and manage increasing labor costs.
