Global AI Weekly
Issue number: 114 | Tuesday, September 9, 2025
Highlights
OpenAI to launch its first AI chip in 2026 with Broadcom
OpenAI will debut its first in-house AI chip next year in partnership with Broadcom. Unlike cloud rivals, the chip will be used internally to power OpenAI’s own systems rather than sold externally. The move comes as the ChatGPT maker seeks to reduce reliance on Nvidia and diversify its supply chain, after previously exploring collaborations with TSMC and AMD. OpenAI’s push mirrors strategies by Google, Amazon, and Meta, who have already developed custom silicon to meet soaring AI compute demands.
reuters.com
Bill Gates: Within 10 years, AI will replace many doctors and teachers—humans won't be needed 'for most things'
Bill Gates shares his vision for the future, predicting that advancements in AI will transform industries like healthcare and education. He explains how AI could handle many tasks currently performed by doctors and teachers, potentially reducing the need for human involvement in various roles. Gates emphasizes the profound impact this "free intelligence" revolution could have on the way we work and live over the next decade.
cnbc.com
OpenAI announces AI-powered hiring platform to take on LinkedIn
OpenAI is preparing to unveil an AI-powered hiring platform by mid-2026, aiming to revolutionize the way candidates connect with businesses. The platform will leverage AI to streamline the recruitment process, offering a more efficient and tailored approach to job matching. This move positions OpenAI as a competitor to established job networking services like LinkedIn.
techcrunch.com
Family of dead teen say ChatGPT's new parental controls not enough
OpenAI has implemented new parental control features following a tragic incident where a family blamed ChatGPT for influencing their teen's decision to take his own life. Despite these changes, the family argues that the measures fall short in addressing the risks of AI interactions with vulnerable users. They are calling for stricter regulations and oversight to prevent similar incidents in the future.
bbc.comResearch
Supervised learning in DNA neural networks
This research explores how DNA molecules can be programmed to autonomously perform supervised learning, showcasing their ability to classify patterns from molecular inputs and desired responses. By integrating training data directly into molecular concentrations, DNA neural networks processed and learned from examples without external computation. This advancement highlights the potential for molecular systems to execute complex tasks, paving the way for molecular machines with embedded learning capabilities in fields ranging from biomedicine to soft materials.
nature.com
Mixture of Global and Local Experts with Diffusion Transformer for Controllable Face Generation
This paper explores a novel approach to controllable face generation by combining a Mixture of Global and Local Experts with a Diffusion Transformer. It introduces an advanced framework designed to enhance generation quality and controllability, leveraging both global and local context understanding. The method shows promising results in creating realistic and diverse facial imagery while allowing fine-grained control over various facial attributes.
huggingface.co
Map of 600,000 brain cells rewrites the textbook on how the brain makes decisions
A groundbreaking study has revealed that the brain's decision-making process involves activity across a much wider network of brain cells than previously understood. By mapping 600,000 brain cells, researchers discovered a more complex and distributed system at play, challenging traditional models of how decisions are made. This discovery could reshape our understanding of brain function and its underlying mechanisms.
livescience.comVideo
AI Study Hour: gpt-oss
This is the recording of a live stream from Pamela Fox where she experiments with GPT-OSS to see what scenarios it works in, and with what Python libraries and frameworks. GPT-OSS is OoenAI's first open-source model family. With a mixture-of-experts design, it delivers 120B parameters at the compute cost of 40B. Lightweight and efficient, it runs on consumer GPUs, bringing advanced AI within everyone’s reach.
youtube.com
AI in Manufacturing: Predictive Maintenance for ROI & Uptime
AI-powered predictive maintenance is revolutionizing manufacturing, drastically cutting machine downtime by 50% and lowering maintenance costs by 31%. With the ability to predict failures with 85% accuracy and provide warnings up to 12 days in advance, this approach promises a 385% ROI for adopters. The session highlights the tech behind these advancements, such as cloud platforms, edge computing, digital twins, and anomaly detection, while addressing challenges like data integration and organizational hurdles. As the AI-in-maintenance market surges, it’s a must-know for manufacturing leaders looking to stay competitive.
youtube.comArticles
How Grok, ChatGPT, Claude, Perplexity, and Gemini handle your data for AI training
AI chatbots like Grok, ChatGPT, Claude, Perplexity, and Gemini have become essential in daily tasks, but they raise important concerns about data privacy. This content explores how user interactions are handled and whether they're used for AI training, shedding light on what happens when you share prompts or personal files with these platforms. Understanding these practices is vital as AI tools become more integrated into our lives.
digit.in
New AI model predicts which genetic mutations truly drive disease
Scientists at Mount Sinai have developed an AI system that uses machine learning, electronic health records, and routine lab tests to evaluate the likelihood of genetic mutations causing diseases. The system generates "ML penetrance" scores that assess genetic risk on a spectrum instead of a binary outcome. This approach has reclassified the risk level of some mutations, showing that not all previously assumed harmful variants have significant real-world effects, while others have stronger links to disease than initially thought.
sciencedaily.com
How Small Language Models Are Key to Scalable Agentic AI
The article explores the role of small language models in enhancing the scalability of agentic AI, a transformative force in automation and digital productivity. It highlights how these compact models, with their efficiency and adaptability, are becoming essential for enterprises and developers aiming to optimize AI systems. By focusing on smaller models, industries can achieve more cost-effective and scalable solutions without compromising performance.
developer.nvidia.com
GPT-5 Is Doing Something Absolutely Bizarre
GPT-5’s writing might seem odd and nonsensical to human readers, but it turns out other chatbots interpret it in unexpected ways. The quirks in its output reveal some fascinating interactions between AI systems, showcasing how different models perceive and process language uniquely. It’s a glimpse into the unpredictable nature of AI communication and the complexities of machine interpretation.
futurism.comUpcoming Events
AgentCon 2025
After an exciting set of AgentCon events before the summer holidays, AgentCon is happening in more locations during the second half of the year. The AI Agents World Tour is a series of one-day conferences for developers focused on autonomous AI agents. Spanning cities like San Francisco and Singapore, these events gather top engineers, researchers, and creators shaping the future of AI technologies. It’s an opportunity to connect, learn, and innovate in the fast-evolving world of autonomous AI. Check all events and register for an event near you!
globalai.communityCode
Top 7 Small Language Models
Small language models are making a big impact in AI by offering a combination of efficiency, reliable reasoning, factual accuracy, and effective tool usage, all while being widely accessible. This overview highlights the top seven small models that are pushing the boundaries of what compact AI systems can achieve.
kdnuggets.com
MCP Authorization With Spring AI and OAuth2
This guide explains how to implement secure authorization for MCP servers and clients using Spring AI and OAuth2. It walks you through setting up OAuth2 configurations and integrating them with Spring AI to protect your applications effectively. By following the steps, you can enhance your system's security while managing authorization processes seamlessly.
baeldung.com
webup/langgraph-up-react: LangGraph template for a simple ReAct agent, with MCP tools support
This repository offers a LangGraph template designed to create a straightforward ReAct agent. It includes integrated MCP tools support and comes with reliable test suites to ensure stability and functionality. A great starting point for those looking to build and test agents effectively with these features.
github.com
AI converts ASCII game to real-time AI-rendered graphics
A software engineer showcases a fascinating project that uses AI to upgrade an ASCII game renderer into real-time, AI-rendered graphics. The transformation brings an impressive visual leap, enhancing the artistic potential of the game. However, challenges remain with latency and consistency, leaving room for further refinement.
tomshardware.comPodcast
AI Today Podcast
The AI Today Podcast offers practical, no-nonsense discussions on how artificial intelligence is being applied across enterprises, public sectors, and leading tech companies. Featuring insights from thought leaders, experts, and industry professionals, this podcast keeps you informed about real-world AI applications without the hype. Perfect for anyone eager to stay updated on the evolving AI landscape.
open.spotify.com