Sakana AI Announces AI CUDA Engineer System, Can Improve Training Efficiency 10-100x

⚡️ Headlines

🤖 AI

Open Source AI Advocates Launch $10 Million Campaign - The Open Source AI Foundation has initiated a $10 million advertising campaign to promote the benefits of open-source artificial intelligence to policymakers and the public. [Axios].

OpenAI's Weekly Active Users Surpass 400 Million - OpenAI's weekly active users exceeded 400 million in February, demonstrating significant growth despite competition from China's DeepSeek. [Reuters].

NVIDIA Unveils AI Platform for Teaching American Sign Language - NVIDIA, in collaboration with the American Society for Deaf Children, has introduced 'Signs,' an AI-driven platform designed to facilitate the learning of American Sign Language through a validated dataset and interactive tools. [NVIDIA Blog].

Together AI Achieves $3.3 Billion Valuation Amid Growing Demand - Together AI, an AI startup supported by Nvidia, has reached a valuation of $3.3 billion following a $305 million funding round led by General Catalyst, reflecting strong corporate demand for AI solutions. [Bloomberg].

Musubi Raises $5 Million for AI-Powered Content Moderation - Musubi, a startup utilizing AI to moderate online content, has secured $5 million in seed funding to enhance its PolicyAI and AIMod systems, aiming to improve accuracy in content moderation. [NBC Philadelphia].

🦾 Emerging Tech

Financial Institutions Support U.S. Digital Assets Leadership - In a joint letter, financial associations have proposed recommendations to bolster U.S. leadership in digital assets by removing obstacles to bank involvement in the sector. [Bank Policy Institute].

SEC Approves First Yield-Bearing Stablecoin - The U.S. Securities and Exchange Commission has approved Figure Markets' application to launch YLDS, the first interest-bearing stablecoin in the United States, offering a yield of 3.85% to holders. [Cointelegraph].

Figure Introduces Helix for Humanoid Robot Control - Figure has unveiled Helix, a Vision-Language-Action model enabling humanoid robots to perform complex tasks, including collaborative object manipulation and real-time adaptation to new environments. [Figure].

🤳 Social Media

X Faces Potential Ad Deal Challenges - Negotiations between social media platform X and advertising conglomerate Interpublic Group have hinted at potential coercion to increase ad spending on X, with implications for Interpublic's merger with Omnicom Group. [The Wall Street Journal].

YouTube Plans Lower-Priced Ad-Free Subscription Tier - YouTube is preparing to launch a more affordable, ad-free version of its subscription service, YouTube Premium Lite, targeting viewers interested in non-music content. [Bloomberg].

Elon Musk Addresses Community Notes Discrepancy - Elon Musk has pledged to rectify issues with X's Community Notes feature after it contradicted his statements, highlighting challenges in content verification on the platform. [Gizmodo].

🔬 Research

MIT Develops AI System to Predict Protein Fragment Interactions - MIT researchers have created FragFold, a computational method that predicts which protein fragments can bind to or inhibit target proteins, potentially advancing biological research and therapeutic applications. [MIT News].

⚖ Legal

SEC Establishes Unit to Combat Fraud in Crypto and AI - The Securities and Exchange Commission has launched a new unit, the Crypto and AI Taskforce Unit (CETU), focused on protecting investors against fraud involving blockchain technology, cryptocurrencies, and artificial intelligence. [The Block].

🎱 Random

Amazon MGM Studios Gains Creative Control Over James Bond Franchise - Amazon MGM Studios has formed a joint venture with Michael G. Wilson and Barbara Broccoli, granting the studio creative control over the James Bond franchise and marking a new era for the iconic series. [About Amazon].

🔌 Plug-Into-This

Sakana AI's AI CUDA Engineer: Revolutionizing AI Training Efficiency

Sakana AI has unveiled the AI CUDA Engineer, an agentic system designed to automate the creation of highly optimized CUDA kernels, achieving speedups ranging from 10 to 100 times over standard PyTorch operations. This innovation aims to significantly reduce computational costs and time in AI model training and deployment.

• Automated CUDA Kernel Generation: The AI CUDA Engineer translates high-level PyTorch code into efficient CUDA kernels, eliminating the need for manual GPU programming expertise.

• Evolutionary Optimization Techniques: Utilizing evolutionary algorithms, the system iteratively refines kernel performance, employing strategies like 'crossover' operations and maintaining an 'innovation archive' to build upon previous advancements.

• Comprehensive Kernel Library: To date, over 17,000 verified CUDA kernels have been generated, covering a wide array of machine learning operations, and are accessible for further research and development.

• Enhanced Performance Benchmarks: In addition to outperforming native PyTorch implementations, the AI CUDA Engineer has produced kernels up to 5 times faster than existing optimized CUDA kernels commonly used in production environments.

🍣 By automating and optimizing GPU computations, this system not only accelerates AI workflows but also has potential to lower the barrier for accessing high-performance computing resources.

FTC Investigates Tech Censorship

The Federal Trade Commission (FTC) has initiated an inquiry into potential censorship practices by major technology companies, aiming to assess whether these actions violate consumer protection laws and suppress free speech.

• Public Call for Comments: The FTC is soliciting public input until May 21, encouraging individuals who have experienced account bans, shadow banning, or content demonetization to share their experiences.

• Leadership and Intent: FTC Chairman Andrew Ferguson, appointed by President Trump, emphasizes that tech firms should not "bully" users, suggesting that such censorship is "not just un-American" but "potentially illegal."

• Scope of Investigation: The inquiry focuses on understanding how companies determine which content to block, with concerns that unpredictable moderation policies may harm consumers and stifle competition.

• Market Impact: Following the announcement, social media stocks experienced declines, notably Reddit's shares dropping over 9%, reflecting investor apprehension regarding potential regulatory actions.

🔍 The FTC's proactive stance aligns with ongoing efforts by the Trump administration to scrutinize and address perceived biases and censorship by Big Tech platforms, particularly against conservative viewpoints.

Rabbit Showcases AI Agent Demonstration

Rabbit has unveiled a "generalist Android agent" capable of autonomously operating applications on a tablet, a feature initially promised with its R1 device over a year ago. This demonstration utilized a laptop-based prompting system to control actions on an Android tablet, bypassing the R1 hardware.

• Demonstration Details: Engineers directed the AI agent to perform tasks such as locating YouTube videos, retrieving cocktail recipes, managing a grocery list, and playing the puzzle game 2048. The agent successfully executed these tasks, albeit with occasional quirks like improperly formatted messages.

• Technical Approach: The AI agent operates by translating high-level user prompts into specific actions within Android applications, effectively automating complex workflows without manual intervention.

• Development Context: This demonstration builds upon Rabbit's previous release of the Large Action Model (LAM) Playground, a "generalist web agent" introduced last year, showcasing the company's ongoing efforts to enhance AI-driven automation.

🐇 Notably, the R1 device was absent from the demonstration, highlighting Rabbit's shift towards software-based solutions that can function independently of proprietary hardware.

 🆕 Updates

📽️ Daily Demo

🗣️ Discourse