Global AI Weekly
Issue number: 93 | Tuesday, April 15, 2025
Highlights
New open source AI company Deep Cogito releases first models and they're already topping the charts
Deep Cogito, a new open-source AI company, has made an impressive debut with its first lineup of models featuring parameter sizes ranging from 3 billion to 70 billion. These models are already gaining attention for their performance and are rapidly climbing the charts in the AI world. This marks an exciting start for the company as it establishes itself in the competitive AI landscape.
venturebeat.com
Meta's answer to DeepSeek is here: Llama 4 launches with long context Scout and Maverick models, and 2T parameter Behemoth on the way!
Meta has introduced the Llama 4 lineup, featuring the Scout and Maverick models designed for extended context comprehension. Despite competition from DeepSeek R1 and OpenAI o1 in certain areas, the upcoming 2-trillion-parameter Behemoth model positions itself as a strong contender in the AI race. This release highlights Meta's push to innovate and remain a key player in the evolving field of AI technology.
venturebeat.comResearch
Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems
This paper explores the progress and ongoing challenges in the development of foundation agents, focusing on their inspiration from brain-like intelligence and the integration of evolutionary, collaborative, and safety-focused approaches. It highlights the potential of these agents to transform technology while addressing critical issues like alignment and long-term reliability. The discussion emphasizes the need for advancing foundational research to create systems that are both innovative and socially responsible.
arxiv.org
An Empirical Study of GPT-4o Image Generation Capabilities
This paper explores the image generation capabilities of GPT-4, providing insights into its performance and potential applications. It examines how the model processes visual prompts and generates imagery, analyzing strengths, limitations, and opportunities for improvement. A detailed evaluation offers a comprehensive understanding of its current capabilities in the rapidly evolving AI landscape.
huggingface.co
Attribution Graphs: A New Interpretability Method to Trace Internal Reasoning in Claude 3.5 Haiku
Anthropic's latest AI research unveils Attribution Graphs, a groundbreaking interpretability method designed to trace and visualize the internal reasoning processes of Claude 3.5. This approach offers a clearer understanding of how the AI reaches its decisions, improving transparency and trust in machine learning systems. The paper provides insights into advancing interpretability for complex AI models.
marktechpost.comVideo
NVIDIA CEO Jensen Huang's Vision for the Future
In this episode of *Huge Conversations*, NVIDIA CEO Jensen Huang shares his insights on the transformative power of GPUs, AI, and cutting-edge technology that has revolutionized industries from gaming to robotics and beyond. Highlighting NVIDIA's journey from its early days to shaping today's AI boom, Huang discusses the company's bold bets on futuristic innovations, their societal impact, and what lies ahead in the next decade. This engaging interview provides a glimpse into the mind of one of tech's most influential figures, offering valuable perspectives on how we can prepare for the rapidly evolving future.
youtu.be
Global AI Bootcamp 2025 - Keynote
In the Global AI Bootcamp 2025 keynote, experts from the Global AI Community alongside renowned speakers like Scott Hanselman, Guido van Rossum, and Jennifer Marsman discuss the evolving role of AI for developers. They highlight Python’s significance in this transformation and showcase innovative advancements with AI agents. The discussion emphasizes the necessity for developers to embrace AI skills while maintaining ethical considerations, offering insights into the current state and future potential of AI in development.
youtube.com
Farzad Sunavala - Silicon Minds, Human Hearts
This episode of *Silicon Minds Human Hearts* features Farzad Sunavala, Principal Product Manager at Microsoft, discussing the transformative potential of Azure AI Search and Retrieval Augmented Generation (RAG). He explores how integrating GPT models with enterprise data is shaping AI's role in business and the broader implications for future AI agents. Packed with insights on innovation and the evolving relationship between humans and technology, this conversation offers a glimpse into how AI continues to impact our work and daily lives.
youtube.comArticles
Introducing the Llama 4 herd in Azure AI Foundry and Azure Databricks
The first models in the Llama 4 herd are now available in Azure AI Foundry and Azure Databricks, offering developers the tools to create personalized multimodal experiences. These Meta-designed models integrate text and vision tokens into a single model backbone, making them ideal for working with large amounts of unlabeled text, image, and video data. This release represents a significant step forward in AI development and application versatility.
azure.microsoft.com
A nonprofit is using AI agents to raise money for charity
Sage, a nonprofit organization, is conducting an experiment to explore how AI agents can effectively raise funds for charities. The initiative allows these AI-driven agents to choose a charity and use their capabilities to gather donations. This innovative approach aims to test the potential of AI in supporting philanthropic causes while showcasing its practical applications in the nonprofit sector.
techcrunch.com
Agent mode: available to all users and supports MCP
Agent mode is now accessible to everyone and comes with full support for MCP. This update ensures a more seamless and inclusive experience for all users, enhancing functionality and efficiency. Explore the new capabilities and take full advantage of this expanded feature set.
code.visualstudio.com
Introducing AutoRAG: fully managed Retrieval-Augmented Generation on Cloudflare
AutoRAG provides a streamlined solution for building and scaling Retrieval-Augmented Generation (RAG) pipelines, ideal for enhancing AI-driven chatbots and search applications. Leveraging Cloudflare's extensive global network and developer tools, it simplifies the process of creating context-aware AI systems with ease and efficiency.
blog.cloudflare.comCode
GitHub - AzureCosmosDB/CosmosAIGraph: CosmosAIGraph implementation of OmniRAG pattern
This repository focuses on the CosmosAIGraph implementation of the OmniRAG pattern. It is part of the AzureCosmosDB project, allowing developers to explore and contribute to its development with ease. Get hands-on with advanced features and collaborate on enhancing the integration of AI capabilities with CosmosDB.
github.com
GitHub - microsoft/ai-agents-for-beginners: 10 Lessons to Get Started Building AI Agents
This GitHub repository by Microsoft provides 10 beginner-friendly lessons to help you get started with building AI agents. It offers a hands-on learning experience, focusing on foundational concepts and practical applications to guide you through the basics of AI agent development. The resources are designed to be easy to follow, making it accessible for those new to the field.
github.comPodcast
Latent Space: The AI Engineer Podcast
Latent Space: The AI Engineer Podcast is a must-listen for AI enthusiasts, offering insights into the latest developments in Software 3.0 and Foundation Models. With news, paper discussions, and exclusive interviews from industry leaders like OpenAI, Meta, and Databricks, the podcast keeps you ahead on topics like code generation, multimodality, AI agents, and GPU infrastructure. Whether you're seeking in-depth takes on current trends or early introductions to emerging tech, this platform is tailored for AI engineers pushing the boundaries of innovation.
open.spotify.com