Global AI Weekly
Issue number: 88 | Tuesday, March 11, 2025
Highlights
}
In another chess move with Microsoft, OpenAI is pouring $12B into CoreWeave | TechCrunch
OpenAI has struck a major five-year, $11.9 billion deal with CoreWeave, a cloud service provider known for its GPU power. This partnership is set to bolster OpenAI’s infrastructure, giving it the computing muscle needed to handle large-scale AI operations. The move further emphasizes OpenAI's commitment to expanding its capabilities alongside its ongoing collaboration with Microsoft.
techcrunch.com
}
Who bought this smoked salmon? How ‘AI agents’ will change the internet (and shopping lists)
Autonomous digital assistants are stepping into everyday tasks like shopping, but their decision-making can lead to unexpected surprises. These AI agents operate independently, raising questions about trust, control, and how they’ll reshape online interactions and consumer habits. They promise convenience but require careful oversight to ensure they align with user preferences.
theguardian.comResearch
}
Intuitive physics understanding emerges from self-supervised pretraining on natural videos
This paper explores how self-supervised pretraining on natural videos can lead to an intuitive understanding of physics. The researchers demonstrate that models trained in this way can predict physical outcomes and interactions without explicit supervision, showcasing a promising approach for developing AI systems with more human-like reasoning about the physical world. The findings highlight the potential of leveraging visual experiences to enhance machine learning capabilities.
arxiv.org
}
Marco: Configurable Graph-Based Task Solving and Multi-AI Agents Framework for Hardware Design | Research
Hardware design is becoming increasingly complex, leading to longer times for optimizing performance, power, area, and cost (PPAC) across various stages like synthesis and verification. This research introduces Marco, a framework that uses configurable graph-based task solving and multi-AI agents to address these challenges. By leveraging the capabilities of large language models (LLMs), Marco aims to enhance efficiency and unlock new possibilities in hardware design workflows.
research.nvidia.comVideo
}
Keynote: AI without the BS, for humans - Scott Hanselman - NDC London 2025
In this insightful keynote from NDC London 2025, Scott Hanselman tackles the practicality of AI without the hype. He discusses the capabilities of very large and small language models, the role of NPUs in everyday devices, and how small models can save power and money. Scott shares real-world examples of using AI to boost productivity while steering clear of empty promises, offering a grounded perspective on its value in our daily lives.
youtube.com
}
Model Mondays – Hands On With GitHub Models
Model Mondays kicks off with an exciting episode that introduces the series and outlines the AI models set to take the spotlight this season. This week, the focus is on GitHub Models, showcasing how to explore and get hands-on experience using just a GitHub account. Perfect for beginners and enthusiasts, this episode provides practical insights and steps to start experimenting with AI models today.
youtube.comArticles
}
DeepSeek isn’t taking VC money yet — here are 3 reasons why | TechCrunch
DeepSeek’s founder Liang Wenfeng explains why the company is holding off on venture capital funding for now. The focus remains on building a solid product, maintaining control over the company’s vision, and avoiding the pressure to grow at unsustainable speeds. This strategy sets DeepSeek apart in an industry where many rush to secure early funding.
techcrunch.com
}
Are AI-generated video games really on the horizon?
Microsoft and Google have unveiled groundbreaking generative AI models capable of simulating video game worlds, hinting at the potential future of game creation. These models showcase the ability to generate dynamic environments and game elements but come with significant challenges and limitations, such as a lack of consistency and control. While promising, it’s clear there’s still a long way to go before AI-generated games become a reality.
theguardian.com
}
The Model is the Product
This content explores how older datasets can be revitalized and leveraged with modern machine learning models to create innovative solutions. It highlights the potential to extract new value and insights by combining "vintage" data with the latest advancements in AI and data modeling. The focus shifts from just data collection to making the models themselves a central product.
vintagedata.orgUpcoming Events
}
Global AI Bootcamp - Week 2
The Global AI Bootcamp is an exciting event for developers and AI enthusiasts to elevate their skills and connect with a global community. This annual session offers immersive learning experiences paired with hands-on activities to deepen your knowledge in Artificial Intelligence. It's the perfect way to explore the latest in AI while growing as a tech professional.
linkedin.comCode
}
GitHub - microsoft/rag-time: RAG Time: A 5-week Learning Journey to Mastering RAG
RAG Time is a comprehensive 5-week learning program designed to help users master Retrieval-Augmented Generation (RAG) techniques. It provides structured lessons, practical exercises, and resources to build a strong understanding of RAG, enabling participants to effectively implement and utilize this powerful approach. Perfect for learners of all levels, this journey blends theory with hands-on applications for a well-rounded learning experience.
github.com