Global AI Weekly
Issue number: 106 | Tuesday, July 15, 2025
Highlights
Unless users take action, Android will let Gemini access third-party apps
Starting Monday, July 7, Android devices will automatically grant Gemini access to certain third-party apps unless users actively adjust their settings. This change emphasizes the importance of reviewing app permissions to ensure data privacy and control. Android users are encouraged to take action to tailor their app settings according to personal preferences and security needs.
arstechnica.com
Satya Nadella announces BioEmu, Microsoft’s AI to fast-track drug discovery
Satya Nadella has introduced BioEmu, Microsoft's latest AI-focused initiative designed to revolutionize drug discovery by making it faster and more efficient. This cutting-edge technology leverages artificial intelligence to analyze and streamline complex biological data, offering significant advancements for the pharmaceutical and healthcare industries. BioEmu showcases Microsoft’s commitment to using AI for impactful solutions in critical fields like medicine.
livemint.com
OpenAI’s open language model is imminent
OpenAI is gearing up to unveil its latest open language model, with the release potentially just days away. Sources close to the company hint at an exciting development that could arrive as soon as next week. Stay tuned for what promises to be a significant step forward in AI innovation.
theverge.com
Nvidia-backed Perplexity launches AI-powered browser to take on Google Chrome
Nvidia-backed Perplexity AI has unveiled Comet, a privacy-focused, AI-powered browser aimed at challenging Google's Chrome. Comet offers conversational task execution, local data storage (no personal model training), and is currently available to Perplexity Max subscribers at $200/month. The browser integrates smart assistants for tasks like product comparison, meeting booking, and content summarization. Backed by investors like Jeff Bezos and SoftBank, Perplexity is positioning itself in the competitive AI browser space while addressing criticism from publishers with a new partnership program.
reuters.comResearch
Rethinking Verification for LLM Code Generation: From Generation to Testing
This paper explores a fresh perspective on improving the reliability of code generation by Large Language Models (LLMs). It focuses on shifting the emphasis from merely generating code to rigorously testing it, aiming for higher accuracy and practicality in real-world applications. The discussion emphasizes the importance of robust verification methods to enhance the quality of AI-driven code development.
huggingface.coArticles
Introducing Docker Hub MCP Server
Discover how the Docker Hub MCP Server enhances your AI workflows by making container image discovery smarter and more intuitive. With features like natural language repository management, this tool takes the hassle out of finding and managing container images. It’s designed to streamline processes and make working with Docker Hub more efficient for developers and AI enthusiasts alike.
docker.com
How to Fine-Tune Small Language Models to Think with Reinforcement Learning
This guide provides a step-by-step walkthrough for fine-tuning small language models using reinforcement learning techniques in PyTorch. It covers training GRPO reasoning models from scratch, complete with code examples and visual explanations. Perfect for those looking to enhance model reasoning capabilities with hands-on methods.
towardsdatascience.com
Hugging Face opens up orders for its Reachy Mini desktop robots
Hugging Face has announced the launch of two versions of its Reachy Mini desktop robot, designed specifically for creating and coding AI-based applications. This compact and versatile robot aims to provide users, from beginners to advanced developers, with a hands-on approach to exploring AI technologies. Orders are now open, offering a new tool for innovation in the AI space.
techcrunch.com
AWS is launching an AI agent marketplace next week with Anthropic as a partner
Amazon Web Services is introducing an AI agent marketplace that will allow developers to find and integrate various AI models and tools more easily. As part of this initiative, AWS has teamed up with Anthropic, known for its work in AI safety and advanced model development. This partnership aims to provide businesses with diverse and reliable AI solutions through the new platform. The marketplace is set to launch next week, marking a significant step in AWS’s AI strategy.
techcrunch.comCode
10 GitHub Repositories for Mastering Agents and MCPs
Discover 10 GitHub repositories packed with resources to help you master agentic AI applications and multi-agent collaboration protocols (MCPs). Explore free tutorials, guides, example code, research papers, and projects that offer practical insights and tools for building your own AI agents. Whether you're a beginner or looking to enhance your skills, these repositories provide everything you need to dive into the world of intelligent agents.
kdnuggets.com
GitHub Copilot coding agent now uses one premium request per session
GitHub Copilot's coding agent is empowering developers to efficiently address tech debt, resolve bugs, and build new features with ease. By assigning tasks to the agent, it simplifies workflows and enhances productivity. The updated model now uses one premium request per session, streamlining the experience even further.
github.blog
A Code Implementation for Designing Intelligent Multi-Agent Workflows with the BeeAI Framework
This content introduces a practical code implementation using the BeeAI Framework to design intelligent multi-agent workflows. It showcases how the framework facilitates collaboration between multiple agents to achieve complex tasks efficiently. The implementation is explained clearly, making it accessible for developers who want to explore or build AI-driven multi-agent systems.
marktechpost.com
Building a High-Performance Parallel LLM Pipeline Using Weight Optimization, KV Cache, SDPA, and Kubernetes
This article explores how to build a high-performance pipeline for large language models (LLMs) by leveraging techniques like weight optimization, KV cache, and SDPA, all while utilizing Kubernetes for scalability. It highlights best practices for efficient deployments and optimizing resource usage when accessing LLMs via API. Ideal for those working on agents, RAG bots, or LLM apps, it offers practical insights to enhance performance and scalability.
levelup.gitconnected.comPodcast
Women And AI
Women And AI is a community dedicated to empowering women in the field of artificial intelligence. It brings together AI enthusiasts at all levels—from students to seasoned experts—to share knowledge, collaborate, and support one another. The initiative emphasizes the importance of diverse perspectives in creating innovative and inclusive AI solutions while fostering mentorship and growth for a brighter future in technology.
open.spotify.com