March 10 – March 17, 2026 1. Mistral Releases "Mistral Small 4" (Apache 2.0) While everyone was distracted by GPT-5.4, Mistral dropped a massive 119B parameter model yesterday (March 16) that changes the math for enterprise AI. Key points - 40% Latency Reduction: Optimized for "Small" hardware but performs at "Large" model levels. - Fully Multimodal: Native support for text-to-image and document understanding built-in. - Apache 2.0: Unlike GPT-5.4, you can host this yourself, modify it, and own your data. Why it matters It provides a high-performance alternative to OpenAI for companies that cannot send their sensitive data to a 3rd party cloud but still need "Agent-grade" reasoning. 2. OpenAI "Sora 1" Sunset & Sora 2 Rollout On March 13, OpenAI officially retired the original Sora 1 in favor of Sora 2, which is now being integrated directly into the ChatGPT interface. Key points - Native 4K Generation: Clips have moved from 720p to native 4K with significantly better physics. - Audio-to-Video Sync: You can now upload a voice track, and Sora will generate a character that lip-syncs to it perfectly. - Pro Tier Exclusive: Access is being tightened; "Sora 2" now requires a Pro or Team subscription for high-resolution output. Why it matters The era of "one-off AI clips" is over. Sora 2 is designed to be a professional asset generator for marketers and YouTubers, not just a playground toy. 3. Apple Intelligence: "Visual Screen Intelligence" Beta Apple quietly expanded its AI beta on March 13, adding the feature we’ve been waiting for: Cross-App On-Screen Actions. Key points - Contextual "See & Do": Siri can now "see" a tracking number in a text and automatically open the delivery app to track it. - Short-Term Memory: You can ask "What was that address Eric sent?" and it will search across Mail, WhatsApp, and Notes simultaneously. - Private Cloud Compute: All complex processing happens on Apple’s "black box" servers, keeping your data private. Why it matters This is the "Agent" experience for the masses. It doesn't require a prompt; it just understands what you are doing on your phone.

This week marks a major turning point. We are seeing the birth of "AI Teams" and "Computer Use" capabilities that transform AI from a writing assistant into a digital worker. Here are the 7 most important updates from the last week. 1. OpenAI Releases GPT-5.4 & GPT-5.4 Pro OpenAI launched its most capable model yet, GPT-5.4, integrated directly into ChatGPT and the API. Key Points • Native Computer Use: The first mainline model that can "see" and interact with your desktop, mouse, and keyboard. • 1 Million Token Context: Massive memory for analyzing entire codebases or long legal documents. • Upfront Planning: Shows you its "Thinking Plan" before it generates, allowing for mid-response course correction. Why it matters This transforms ChatGPT from a chatbot into a digital worker that can actually perform tasks inside your software autonomously. 2. Google Launches Gemini Embedding 2 Google released its first natively multimodal embedding model for developers via the Gemini API and Vertex AI. Key Points • Unified Understanding: Maps text, image, video, audio, and documents into a single shared vector space. • Interleaved Input: Processes combinations like "image + text" in one request for richer search context. • Matryoshka Scaling: Allows developers to shrink model size for speed without starting from scratch. Why it matters It makes it significantly easier to build "memory" systems for your AI that understand your training videos, PDFs, and meeting recordings all in one place. 3. Gemini Testing "Multi-Agent Planning" Google is testing a new coordination feature in Gemini Business to help manage complex projects. Key Points • Specialist Identification: Automatically finds the best internal AI "agents" for a specific job. • Delegation Plans: Creates a step-by-step plan to distribute work among multiple AI agents. • Workspace Orchestration: Designed to live inside Docs, Sheets, and Drive to manage multi-app workflows. Why it matters We are moving toward "AI Teams" where a central brain manages a group of specialized agents to complete projects for you.

Welcome back to the show! We hear a lot about students using AI to write papers and study, but what happens when the professors start bringing AI into the classroom? Today, we are diving into a fascinating new report from Anthropic. They analyzed around 74,000 anonymized conversations from higher education professionals to see exactly how educators are using Claude. It turns out, educators aren't just using it as a basic chatbot. They are acting as creators, using Claude Artifacts to build fully functional, interactive educational tools—think chemistry simulations, web-based escape rooms, and data visualisation dashboards. But it's not all smooth sailing. While many teachers love using AI as a collaborative thought partner to design curricula and automate tedious admin work, there is a massive, contentious debate brewing over using AI to grade student work. So, are robots taking over the grading pen? And how is AI forcing professors to completely rethink the traditional research paper? Grab your notebooks, because we're getting into it right now! Podcast

Free Lifetime Access to Prompt Prodigy If you join ClassAi today, you will receive lifetime access to Prompt Prodigy. Prompt Prodigy is a structured, interactive tool designed to help you: • Write clearer prompts• Improve AI outputs• Reduce time spent rewriting• Build confidence using AI in teaching. It guides you step by step. aIt gives feedback, it helps you improve in real time. This is not a PDF it,s an interactive learning tool. Why am I doing this? Because prompt quality is the difference between average AI use and professional AI use. If you are a teacher or educator and want to: • Save time• Improve planning• Mark faster• Build AI confidence This is for you.

If you work in education, these are the AI skills to focus on: 1. Prompt writing 2. AI literacy 3. Critical thinking with AI 4. AI for productivity 5. AI assisted assessment 6. Data awareness 7. Multimodal AI use 8. AI ethics 9. Workflow integration 10. Teaching AI skills to students Build skills, not dependency. Which two are you working on right now?