Global AI Weekly

Issue number: 105 | Tuesday, July 8, 2025

Highlights

Google faces EU antitrust complaint over AI Overviews

Google faces EU antitrust complaint over AI Overviews

The Independent Publishers Alliance has submitted an antitrust complaint to the European Commission, raising concerns about Google’s AI Overviews. They argue that this feature takes content from publishers without proper compensation, potentially harming their businesses. The complaint highlights broader issues about the balance of power between tech giants and independent publishers.

techcrunch.com

Elon Musk confirms xAI is buying an overseas power plant and shipping the whole thing to the U.S. to power its new data center

Elon Musk confirms xAI is buying an overseas power plant and shipping the whole thing to the U.S. to power its new data center

Elon Musk's xAI is making bold moves to support its ambitious AI projects by acquiring and importing an entire power plant to the U.S. This massive effort aims to power a cutting-edge data center that will house one million AI GPUs and consume up to 2 gigawatts of electricity—enough to supply energy to 1.9 million homes. The scale underscores Musk's commitment to advancing AI development, while also highlighting the immense energy needs of such ventures.

tomshardware.com

Amazon Project Rainier: AI supercomputer for Anthropic

Amazon Project Rainier: AI supercomputer for Anthropic

Amazon has unveiled Project Rainier, an AI supercomputer designed to support Anthropic, a safety-focused AI company. This cutting-edge infrastructure aims to boost the development of advanced AI systems, showcasing Amazon's commitment to enhancing AI technology. With its high-performance capabilities, Rainier is poised to play a pivotal role in pushing the boundaries of AI innovation.

theregister.com

Research

LLMs Get Lost In Multi-Turn Conversation

LLMs Get Lost In Multi-Turn Conversation

This abstract highlights the challenges large language models (LLMs) face during multi-turn conversations, with a focus on their tendency to lose conversational context over time. It examines how these models struggle to maintain consistency and accuracy as interactions grow longer, presenting a deeper understanding of their limitations in handling complex, multi-step dialogue. The paper also explores potential approaches to mitigate these issues, aiming to improve their conversational capabilities.

arxiv.org

Reinforcement Pre-Training

Reinforcement Pre-Training

This paper introduces the concept of Reinforcement Pre-Training (RPT), a novel approach that combines reinforcement learning with pre-training techniques to enhance decision-making tasks. It explores how RPT can improve learning efficiency and adaptability across various environments, merging the strengths of both paradigms. The method is applied to several benchmarks, demonstrating promising results and practical implications for advancing AI systems.

arxiv.org

Video

Humans in AI - Chris Lloyd Jones

Humans in AI - Chris Lloyd Jones

In this episode of *Humans in AI*, Chris Lloyd Jones, a software expert from Seattle, discusses creating cutting-edge software-as-a-service solutions for artificial intelligence. He shares insights on the gradual yet impactful evolution of AI at scale, the swift disruptions it may cause in various industries, and the exciting new job opportunities it will generate. Chris also talks about the dynamic ways humans are poised to adapt to these transformative changes.

youtube.com

What's New in Azure AI Foundry & AI Center of Excellence

What's New in Azure AI Foundry & AI Center of Excellence

Catch the latest updates on Azure AI Foundry with John Maeda in this exciting session from Monday, June 30. Plus, Cyril Belikoff joins Seth to explore how the Azure Essentials AI Center of Excellence can help drive impactful AI transformations at scale. Don't miss this insightful look into AI advancements and guidance!

youtube.com

Articles

Could Google’s Veo 3 be the start of playable world models?

Could Google’s Veo 3 be the start of playable world models?

Demis Hassabis, CEO of Google DeepMind, suggested that Veo 3, Google’s cutting-edge video generating model, might find applications in the gaming industry. The technology has the potential to create playable world models, opening doors to more immersive and dynamic gaming experiences. While details are scarce, this could mark a significant step forward in how AI contributes to interactive entertainment.

techcrunch.com

How IBM helps Wimbledon use generative AI to drive personalised fan engagement

How IBM helps Wimbledon use generative AI to drive personalised fan engagement

IBM is using generative AI technology to enhance the fan experience at Wimbledon by delivering personalized and engaging content. This innovative approach helps fans stay connected to the tournament through tailored updates, highlights, and interactive features. By integrating AI, Wimbledon is creating deeper connections with its global audience while staying at the forefront of tech-driven sports experiences.

ibm.com

Efficient GPT-4V level multimodal large language model for deployment on edge devices

Efficient GPT-4V level multimodal large language model for deployment on edge devices

This week's spotlight is on MiniCPM-V, a breakthrough in efficient multimodal large language models designed for edge devices. These lightweight models deliver impressive performance, surpassing GPT-4V and other leading models across benchmarks, while running efficiently on mobile phones. With support for over 30 languages, robust image processing, and reduced computational demands, they make advanced AI more accessible for mobile, offline, and privacy-sensitive applications. This innovation highlights a pivotal shift toward enabling powerful multimodal intelligence directly on consumer hardware.

nature.com

NotebookLM started as a Google Labs experiment, so I tested other Labs projects to see how they stack up

NotebookLM started as a Google Labs experiment, so I tested other Labs projects to see how they stack up

Google Labs is rolling out experimental projects like NotebookLM, and it's catching attention for its innovation. This article takes a closer look at NotebookLM's capabilities and dives into other intriguing Labs experiments to see how they measure up. From AI tools to creative tech solutions, it explores the potential and surprises within these projects.

xda-developers.com

Code

I Made Cursor + AI Write Perfect SQL. Here's the Exact Setup

I Made Cursor + AI Write Perfect SQL. Here's the Exact Setup

Learn how to streamline the process of generating accurate SQL queries with AI using a precise Cursor and MotherDuck setup. This guide walks you through a step-by-step approach to ensure the AI produces working SQL scripts right from the start, saving time and effort on debugging. Perfect for anyone looking to enhance their workflow with efficient AI-driven database queries.

motherduck.com

Open Source AI Editor: First Milestone

Open Source AI Editor: First Milestone

GitHub has announced the open-sourcing of the Copilot Chat extension as a major step toward transforming VS Code into an open source AI-driven editor. This move marks a significant milestone in their efforts to make AI tools more accessible and customizable for developers. It’s a step forward in fostering collaboration and innovation within the developer community.

code.visualstudio.com

Podcast

The AI Daily Brief (Formerly The AI Breakdown): Artificial Intelligence News and Analysis

The AI Daily Brief (Formerly The AI Breakdown): Artificial Intelligence News and Analysis

The AI Daily Brief offers a daily news analysis on everything related to artificial intelligence. Hosted by NLW, the show explores topics ranging from creative advancements with tools like Midjourney and ChatGPT to the potential impact AI could have on industries and work. It also touches on deeper questions around ethics, alignment, and the risks of advanced AI systems.

open.spotify.com

>