Unlock the full potential of your workday with cutting-edge AI strategies and actionable insights, empowering you to achieve unparalleled excellence in the future of work. Download the free guide today!

Today:

• Hugging Face, NVIDIA Boost Open-Source AI Robotics

• OpenAI Acquires High-Value Chat.com Domain

• Perplexity Raises $500M, Hits $9B

• Microsoft Enters Multi-AI Agent Race

Hugging Face, NVIDIA Boost Open-Source AI Robotics

Hugging Face and NVIDIA are teaming up to boost robotics research with open-source tools that make it easier for developers to build advanced AI robots. Through Hugging Face's LeRobot platform and NVIDIA's robotics tech, they provide resources for faster, more collaborative robot development. 

Using NVIDIA's Isaac Lab, researchers can simulate real-world interactions in virtual environments to train robots effectively, saving time and resources. The project encourages sharing data, models, and tools across the community, helping speed up advancements in industries like healthcare, manufacturing, and logistics. This partnership aims to drive widespread adoption of intelligent robotics.

OpenAI Acquires High-Value Chat.com Domain

OpenAI has acquired the domain Chat.com, which now redirects to its ChatGPT platform, enhancing its portfolio of valuable domain names. Chat.com, first registered in 1996, was previously purchased by HubSpot’s co-founder for $15.5 million, one of the highest publicized domain sales. 

OpenAI hasn't disclosed the acquisition price, and the domain change doesn't appear to signal a rebranding effort. This acquisition aligns with OpenAI’s strategy of broadening accessibility to its AI services, positioning ChatGPT on a memorable, high-profile domain without significant branding shifts.

Perplexity Raises $500M, Hits $9B

AI search startup Perplexity is set to triple its valuation to $9 billion with a new $500 million funding round, led by Institutional Venture Partners, marking its fourth round this year. This rapid growth highlights strong investor interest in artificial intelligence, particularly in AI-powered search technology. 

CEO Aravind Srinivas is steering Perplexity through this expansion, showing the continued momentum of AI startups in the tech investment landscape. Perplexity’s valuation increase underscores the industry's high expectations for AI-driven solutions in the search market.

Microsoft Enters Multi-AI Agent Race

Microsoft’s new multi-AI agent system, Magnetic-One, is designed to automate complex tasks for businesses, reducing human intervention. Built on Microsoft’s open-source AutoGen framework, Magnetic-One operates through an Orchestrator agent that directs specialized sub-agents: WebSurfer, FileSurfer, Coder, and Computer Terminal, each handling tasks like browsing, file management, coding, and running programs. 

Microsoft advises caution, recommending isolation and monitoring to prevent risks. This system competes with similar tools from Salesforce and IBM, signaling Microsoft’s push into enterprise-level AI solutions aimed at improving efficiency in completing intricate, traditionally human tasks.

🧠RESEARCH

Hunyuan-Large: An Open-Source MoE Model with 52 Billion Activated Parameters by Tencent

Hunyuan-Large is Tencent's open-source mixture of experts (MoE) model, featuring 389 billion parameters and 52 billion active parameters. It can process up to 256,000 tokens, offering advanced performance in various tasks like language understanding, logic, coding, and more. Hunyuan-Large surpasses LLama3.1-70B and matches LLama3.1-405B despite its smaller size. Key innovations include large-scale synthetic data, mixed expert routing, key-value cache compression, and tailored learning rates for experts.

gsplat: An Open-Source Library for Gaussian Splatting

gsplat is an open-source library designed to improve Gaussian Splatting, a method for 3D image processing. Built for compatibility with PyTorch, it enhances speed, memory efficiency, and model training time by up to 10%. Actively maintained on GitHub, gsplat encourages community collaboration and offers optimized tools for advanced 3D visual projects.

Photon: Federated LLM Pre-Training

Photon is a federated system enabling large language model (LLM) pre-training across decentralized GPUs with minimal communication needs. This approach, unlike traditional data center training, uses federated learning to train LLMs globally, achieving faster and more efficient performance. Photon reduces communication by up to 512 times, trains models up to 7B parameters, and reaches better accuracy than centralized methods, marking a significant advancement in economical, large-scale model training.

DeeR-VLA: Dynamic Inference of Multimodal Large Language Models for Efficient Robot Execution

DeeR-VLA introduces a framework to improve efficiency in multimodal language models (MLLMs) for robots, adjusting model size dynamically based on task needs. This approach, DeeR, employs a multi-exit structure, allowing the model to end processing early when sufficient understanding is reached, reducing computation and memory demands. Tested on the CALVIN benchmark for robot tasks, DeeR cuts computational costs by 5.2-6.5 times and GPU memory by 2-6 times while preserving performance, making it practical for real-world robotic applications with limited resources.

HtmlRAG: HTML is Better Than Plain Text for Modeling Retrieved Knowledge in RAG Systems

HtmlRAG is a new approach for Retrieval-Augmented Generation (RAG) that improves how knowledge from the Web is used in language models. Instead of converting HTML to plain text, HtmlRAG keeps the HTML format, retaining valuable structural information like headings and tables. To handle extra HTML content (e.g., tags, JavaScript), the researchers introduced methods for HTML cleaning, compression, and selective pruning. Tests on question-answering datasets show that HtmlRAG outperforms traditional plain-text RAG, making it a promising solution for enhanced knowledge representation in RAG systems.

🛠️TOP TOOLS

My Audio Journal - Your daily workout for the mind

Kiwi Fitness - Step-by-step strength workouts tailored for you.

Sourcely - Find Academic Sources in Seconds.

Peech -  AI-powered platform automates video management, editing, and repurposing.

Nifty - The ultimate project management with AI-powered tools

📲SOCIAL MEDIA

🗞️MORE NEWS

• Perplexity, an AI search engine known for controversial content scraping, launched a real-time election tracker for the 2024 US election. Using strict information sources, it provided accurate results, contrasting with cautious approaches from other AI companies.

• Microsoft is rolling out new AI tools for Windows Insiders in Paint and Notepad. Paint now includes generative fill and erase features, allowing users to edit images intuitively. Notepad introduces an AI-powered rewrite tool for rephrasing, adjusting tone, and modifying text length. 

• AIRIS, an AI developed by SingularityNET and ASI Alliance, is learning Minecraft through self-directed exploration in a 3D world. It navigates complex environments, builds internal maps, and adapts to obstacles, aiming to explore unknown areas without specific instructions. 

• Apple is researching smart glasses with an internal project called "Atlas," involving employee feedback on current products. This signals Apple’s potential entry into the smart glasses market, aiming to compete with Meta in this growing tech category.

• T-Mobile has committed to paying OpenAI approximately $100 million over the next three years to develop a customer service chatbot. This AI-driven platform aims to utilize generative artificial intelligence to handle the majority of customer inquiries by remembering customer-specific data, thereby reducing the need for human intervention.