Sponsored by

Today:

Transform your AI knowledge with The Rundown - the fastest growing AI newsletter in the world

Receive daily insights on cutting-edge AI tools and workflows to boost your productivity, along with the latest and greatest news in the world of AI.

• Trusted by 400,000+ daily readers for the latest in AI, including top industry experts

• Discover AI tools and tutorials that skyrocket efficiency and innovation.

• Stay on top of AI trends and techniques to enhance your work and productivity

OpenAI Backed Robot | THE ROBOTIC REVOLUTION WILL NOT BE TELE-OPERATED (!!!!)

Sam Altman's hinted-at robotic startup has made headlines with a Twitter post by Eric Jang showcasing their robot performing tasks with no human puppeteering, no shortcuts, just one neural net calling the shots. 

They're hiring to expand these capabilities, signaling a promising future for robotics. The venture not only advances AI but also introduces roles like Android operators, blending tech and creativity to train robots, illustrating the evolving job landscape in the AI era.

WATCH THE VIDEO ON YOUTUBE

Sam Altman Seeks Trillions of Dollars to Reshape Business of Chips and AI

OpenAI's CEO Sam Altman aims to revolutionize the global semiconductor industry, seeking trillions of dollars in investment, including from the UAE government, for a project potentially costing up to $7 trillion. The initiative intends to address constraints hindering OpenAI’s growth, such as the scarcity of AI chips. 

Altman's vision underscores the demand for AI infrastructure crucial for economic competitiveness. The proposed investment surpasses the current size of the semiconductor industry and could reshape the tech landscape, challenging market leaders like Nvidia. However, the fundraising faces significant hurdles despite Altman's track record and OpenAI's success with ChatGPT.

THE WALL STREET JOURNAL

Nvidia pursues $30 billion custom chip opportunity with new unit -sources

Nvidia, a major player in AI chips, is launching a new unit to design custom chips for cloud computing and AI. With competitors emerging, Nvidia aims to secure its market share and cater to specific needs. Its dominant position in high-end AI chips has driven stock value up, but demand exceeds supply. To adapt, Nvidia seeks to collaborate with tech giants like Amazon, Meta, and Microsoft to create tailored solutions. 

This move targets a $30 billion market, potentially challenging current leaders like Broadcom and Marvell. Beyond AI, Nvidia eyes expansion into telecom, automotive, and gaming sectors.

REUTERS

ASML Shows Off $380 Million, 165-Ton Machine Behind AI Shift

ASML Holding NV flaunts its newest chip-making behemoth, a $380 million mammoth that tips the scales like two Airbus A320s. Dubbed the High-NA extreme ultraviolet system, it's drawing eyes after Intel Corp. placed orders and received the first one in December, shipping it to an Oregon factory. The media got a peek at this heavyweight champion on Friday. 

With Intel gearing up to produce chips using it by late 2025, ASML's colossal creation signals a significant shift in the tech landscape. This unveiling underscores the relentless pursuit of innovation driving the AI race.

BLOOMBERG

Microsoft hints at the future of AI in Windows with a smarter Copilot

Microsoft is gearing up to enhance Windows 11 with more AI features, teasing a smarter Copilot tool. The update includes an animated Copilot in the taskbar that helps with text summarization and explanation. Testing in Windows 11 Canary builds shows promising results, with future plans to support image analysis and editing. Notepad will also integrate Copilot for deciphering text and code segments. 

Microsoft aims for 2024 to be the "year of the AI PC" and is steadily integrating Copilot across Windows features. This AI push aligns with hardware advancements like NPU support in Windows laptops. Expect more AI enhancements in the upcoming Windows 11 update.

THE VERGE

🧠RESEARCH

More Agents Is All You Need

The study found that using more agents in a simple sampling-and-voting system boosts the performance of large language models (LLMs). This improvement works alongside existing complex methods and is more noticeable with harder tasks. After extensive testing across various benchmarks, the authors show this method's effectiveness and explore the conditions that make it work best. They've shared their code online for public use.

WebLINX: Real-World Website Navigation with Multi-Turn Dialogue

WebLINX is a benchmark for testing digital agents' ability to navigate websites based on user instructions in a conversational manner. It includes 100K interactions over 150 real-world websites, aiming to improve agents' web navigation by focusing on relevant parts of HTML pages. Despite smaller finetuned decoders outperforming large models, the challenge remains in generalizing to new websites. The study emphasizes the need for more adaptive multimodal models.

Multilingual E5 Text Embeddings: A Technical Report

This technical report details the development and evaluation of multilingual E5 text embedding models, launched in mid-2023. It offers three sizes of models to balance between quick processing and quality of embeddings. These models were trained using a contrastive pre-training method on 1 billion multilingual text pairs, then fine-tuned on labeled datasets. A notable introduction is an instruction-tuned embedding model, matching the performance of top English-only models. 

λ-ECLIPSE: Multi-Concept Personalized Text-to-Image Diffusion Models by Leveraging CLIP Latent Space

 λ-ECLIPSE is a model enhancing personalized text-to-image generation by utilizing CLIP's latent space, overcoming challenges like high resource needs and output inconsistency common in previous methods. It leverages a streamlined approach, requiring only 34M parameters and 74 GPU hours for training, significantly less than prior models. λ-ECLIPSE excels in generating images that closely align with user instructions, maintaining high quality with lower computational costs. 

Memory Consolidation Enables Long-Context Video Understanding

The paper introduces a method called memory consolidation for enhancing long-context video understanding with video transformers. By fine-tuning pre-trained transformers to focus on past activations, this approach extends the temporal context significantly without adding computational complexity. The memory-consolidated vision transformer (MC-ViT) achieves new benchmarks in video understanding tasks, outperforming larger models with its efficient learning from longer videos.

🛠️TOP TOOLS

Supadash - provides a streamlined dashboard creation service. It simplifies the process of creating and managing dashboards for data visualization and analytics.

Daydream - AI-powered insights & reporting tool for C-Levels, Finance, & Ops.

Gempages - Shopify app that allows users to create visually appealing and conversion-focused landing pages. 

Firstbase - simplifies the process of launching and running a US-based business.

Waymark - AI video creator that streamlines the process of producing TV, CTV, and digital ads.

📲SOCIAL MEDIA

🗞️MORE NEWS

Disney harnesses AI to drive streaming ad technology

Disney utilizes AI for tailored ads on Disney+ and Hulu, analyzing scenes to match ad mood. "Magic Words" personalize ads, shifting from broad demographics to specific audiences. Beta testing includes Omnicom and five others. Amid TV ad decline, Disney aims for streaming ad innovation. CEO projects streaming as prime ad space. REUTERS

Apple Sets New Record: Acquires 32 AI Startups in 2023, the Highest Among Major-Tech Companies

Apple leads major-tech companies with 32 AI startup acquisitions in 2023, outpacing Google, Meta, and Microsoft. Analysts note Apple's strategic move to bolster AI capabilities for future tech developments. While competitors advance with generative AI, Apple's proactive approach targets emerging trends. Its acquisitions, including Voysis and WaveOne, enhance products like Siri and iPhones. STOCKLYTICS

Instagram might let you use AI to write messages

Instagram could introduce AI-powered message writing, hinted by mobile developer Alessandro Paluzzi's screenshot. While its exact function remains uncertain, it aligns with Meta's existing AI features across platforms. This potential addition reflects ongoing advancements in AI integration within social media. THE VERGE

Minecraft could be the key to creating adaptable AI

Researchers at the University of the Witwatersrand in South Africa developed MinePlanner, a Minecraft-based benchmark test to gauge AI models' general intelligence. Unlike traditional AI training, MinePlanner challenges models to ignore extraneous details, fostering adaptability similar to human problem-solving. Existing AI struggled with complex tasks, highlighting the need for AI capable of independent problem assessment. NEWSCIENTIST

These $349 smart glasses have ‘AI superpowers’ and a comical charging nose

Brilliant Labs unveils Frame smart glasses, priced at $349, offering AI features like visual analysis, web search, and translation. The glasses integrate with the Noa app, leveraging OpenAI for visual analysis and Whisper for translation. Frame's open-source design and hackability set it apart in the smart eyewear market, aiming to provide users with freedom and versatility. THE VERGE

Abu Dhabi AI group G42 sells its China stakes to appease US

Abu Dhabi artificial intelligence group G42 has sold its stakes in Chinese companies, including TikTok owner ByteDance, in an effort to reassure US partners and cut ties with China. 42XFund, the $10bn technology investment arm of G42, divested from all its investments in China, including shares in ByteDance. G42 is aiming to become a global leader in AI and has been working with US partners such as Microsoft and OpenAI, as well as Chinese technology companies. The move to sell Chinese holdings comes as G42 seeks to allay fears that companies linked to the Chinese government could gain access to American AI systems. FINANCIAL TIMES

Meet the Pranksters Behind Goody-2, the World’s ‘Most Responsible’ AI Chatbot

"Goody-2" is an AI chatbot designed to take AI safety measures to an extreme level by refusing every request and providing explanations for doing so. This chatbot's responses, while self-righteous and somewhat ridiculous, aim to highlight the challenges and complexities of implementing responsible AI systems. The creators behind Goody-2, Mike Lacher and Brian Moore, describe their project as a satire of the AI industry's emphasis on safety without reservations. They believe it raises important questions about who determines what constitutes responsible AI and how it should be implemented. While Goody-2 may seem like a joke, it underscores the ongoing issues with ensuring the safety and ethical use of AI technology. WIRED