Global AI Weekly - Issue 93 - Global AI Community

In this issue

Highlights

2 stories

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.

In this issue

Research

3 stories

arxiv.org

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.

huggingface.co

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.

marktechpost.com

Attribution Graphs: A New Interpretability Method to Trace Internal Reasoning in Claude 3.5 Haiku

In this issue

Video

3 stories

youtu.be

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.

youtube.com

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

In this issue

Articles

4 stories

azure.microsoft.com

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.

techcrunch.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.

code.visualstudio.com

Agent mode: available to all users and supports MCP

blog.cloudflare.com

Introducing AutoRAG: fully managed Retrieval-Augmented Generation on Cloudflare

In this issue

Code

2 stories

github.com

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.

In this issue

Podcast

1 story

open.spotify.com

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.