The artificial intelligence (AI) realm witnesses rapid developments around the world, reshaping the world of digital transformation and accelerating the pace of AI adoption. In this weekly newsletter, FollowICT is highlighting the most important news and advancements in AI around the globe.
Nvidia to Manufacture AI Supercomputers in the US
Nvidia is teaming up with manufacturing partners to design and build facilities that will, for the first time, produce complete Nvidia AI supercomputers within the United States.
The company has allocated over 1 million square feet of manufacturing space across Arizona and Texas to build and test its new Blackwell chips and AI supercomputers. Blackwell chip production has already kicked off at TSMC’s facilities in Phoenix, Arizona. Nvidia is also building supercomputer manufacturing plants in Texas, partnering with Foxconn in Houston and Wistron in Dallas. Mass production is expected to scale up over the next 12–15 months.
AI chip and supercomputer supply chains are complex, requiring advanced capabilities in packaging, assembly, and testing. Nvidia is collaborating with Amkor and SPIL to handle packaging and testing in Arizona.
OpenAI Upgrades ChatGPT’s Memory—Again
In an update quietly posted to OpenAI’s support and changelog pages, the company introduced “Memory with Search,” a feature that allows ChatGPT to reference past conversation details—like your favorite foods—when searching the web to give more relevant answers.
This comes on the heels of broader memory upgrades that give ChatGPT access to a full chat history. These enhancements appear aimed at helping ChatGPT stay ahead of competitors like Anthropic’s Claude and Google’s Gemini, which also now offer memory features.
xAI Adds Memory Feature to Grok
Elon Musk’s AI company xAI is gradually bringing its chatbot Grok up to speed with its competitors. On Wednesday, xAI announced that Grok now includes a memory feature, allowing it to recall details from past interactions.
This means Grok can now offer more personalized suggestions—like tailored recommendations—based on prior usage. It’s a clear attempt to compete with ChatGPT and Gemini, both of which already support persistent memory capabilities.
Microsoft Researchers Build AI Model Runs on CPUs
Microsoft researchers have developed the most powerful 1-bit AI model to date. Called BitNet b1.58 2B4T, it’s available under an MIT license and can run efficiently on CPUs, including Apple’s M2 processor.
Bit-level networks are compressed models designed for lightweight devices. By quantizing the model’s weights, they drastically reduce the number of bits needed—making them faster and more memory-efficient.
Capsule Raises $12M to Build AI-Powered Video Editing Tool for Brands
AI video platform Capsule announced a $12 million Series A funding round to upgrade its AI assistant for marketing, media, and sales teams.
The upcoming version of the video editor will include real-time collaboration, AI-powered suggestions, and a co-pilot feature that helps brands craft engaging video stories.
Users will be able to prompt the tool with requests like “make a short video from our latest event,” and the AI will suggest everything—from clips to transitions, titles, and graphics—making content creation faster and easier.
Claude Can Now Read Your Gmail
Anthropic announced that its chatbot Claude now integrates with Google Workspace, allowing it to access Gmail, Google Calendar, and Google Docs.
This integration is currently rolling out in beta for users on the Max, Team, Enterprise, and Pro plans. Admins must enable the feature for their teams before connecting Claude to their Google accounts.
Bill Gates Says AI Could Solve Teaching, Healthcare Shortages
In a recent episode of the “People by WTF” podcast, Bill Gates shared that AI could help close the talent gap in critical fields like education and medicine.
“AI will come in and provide medical intelligence. There won’t be a shortage,” Gates said. He emphasized that countries like India and many in Africa still face a shortage of qualified medical professionals—and the US isn’t far behind. A report from the Association of American Medical Colleges projected a shortage of up to 86,000 doctors in the US by 2036.
New ChatGPT Image Feature Sparks Privacy Concerns
OpenAI launched new models, “o3” and “o4-mini”, that can analyze images with impressive precision—even blurry or distorted ones.
These models can crop, rotate, and zoom into images to identify specific details. Users discovered that o3, in particular, is surprisingly good at deducing locations, landmarks, and even local businesses like restaurants and bars, all from subtle visual clues—without relying on image metadata or past conversations.
Tests show that even GPT-4o, an earlier model, can often reach the same conclusions, sometimes even faster. Still, o3 showed some unique advantages in at least one test case. The rise of such powerful visual reasoning models raises new privacy and ethical questions about AI’s growing ability to identify real-world locations.
OpenAI Launches New GPT-4.1 Models
On Monday, OpenAI unveiled a new family of advanced AI models: GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano.
These new multimodal models, available through OpenAI’s API (but not ChatGPT), are said to outperform GPT-4o and GPT-4o mini across the board. They can handle up to 1 million tokens of input—including text, images, and embedded video—making them highly capable across a range of applications.
Google Launches Arabic-Supported Video Generator “Veo 2”
Google’s advanced video-generation model, Veo 2, is now available to Gemini Advanced users in multiple languages, including Arabic and English. Initially unveiled last year, Veo 2 allows users in MENA and worldwide to create and share high-definition, cinematic videos using AI.
Designed with an understanding of real-world physics and human motion, Veo 2 produces highly detailed visuals and vibrant scenes, pushing the boundaries of what AI video generation can do.
AI Basics: What Is Machine Learning?
Machine Learning (ML) is a subfield of AI that enables systems to learn from data without being explicitly programmed.
Instead of hardcoding instructions, we feed the system large datasets so it can recognize patterns and make predictions. The more data it sees, the better it performs.
Types of ML:
• Supervised Learning: Trains on labeled data (e.g., emails marked as “spam” or “not spam”).
• Unsupervised Learning: Finds patterns in unlabeled data (e.g., grouping customers by behavior).
• Reinforcement Learning: Learns through feedback and rewards—often used in robotics or games like chess.
