💼 OpenAI to merge ChatGPT's Chat and Work tabs 💼
A single ChatGPT workspace simplifies access and shared usage limits.

OpenAI is preparing to merge ChatGPT's Chat and Work tabs into a single experience, ending the split between casual conversations and task-oriented workflows. Instead of choosing a mode before starting, users will land in one unified interface where everyday questions and longer work projects live side by side. Codex, the company's coding agent, stays as a separate product but will draw from the same usage pool, meaning limits are shared across ChatGPT and Codex rather than tracked independently. The change, reported by TechCrunch, signals OpenAI's push to make ChatGPT feel like one coherent workplace tool, a clear signal of competitive pressure from rival AI workspaces. For heavy users, the shared pool means planning usage across chat and coding more carefully.

🏫 Google won't give schools Search without AI 🏫
New K-12 defaults tie Search access to Gemini AI tools.

Under new K-12 settings, Google will switch Search, Gemini and Gemini Notebook off by default for schools, but it will not offer an AI-free version of Search. In practice, administrators face a binary choice: enable Google's AI features, or lose access to Google Search entirely across student accounts. The policy, outlined on Google's education blog as part of its "AI learning era" updates, has drawn criticism from former FTC chair Lina Khan. For schools wary of generative AI in classrooms, the options are narrowing fast. The move carries real regulation risk around bundling Search with AI, and could draw fresh antitrust attention to Google's education business.

🛰️ NVIDIA Cosmos adds LiDAR world-model simulation 🛰️
Cosmos integration brings video and LiDAR simulation to robotics.

NVIDIA has expanded its Cosmos framework with new capabilities for training, fine-tuning and reinforcement learning of physical-AI world models. World models let robots and autonomous vehicles learn inside simulated environments that predict how the real world reacts to their actions. The headline addition is synchronized video and LiDAR simulation, so developers can generate matching camera and sensor data for the same scene instead of stitching sources together. That matters for self-driving and robotics teams, who rely on multi-sensor data to test edge cases that are rare or dangerous to capture on real roads. The update deepens NVIDIA's grip on the physical-AI developer stack, adding competitive pressure on rival simulation platforms. Details and guides are available on the NVIDIA Developer blog.

INTERESTING TO KNOW

📺️ Netflix tests AI Discover tab on TV app 📺️

Netflix is testing AI Discover, a new tab on its TV app that helps viewers pick something to watch based on their mood. Users start from AI-generated themes like "Scare Me" or "Quick Watches," then narrow the options from there. The feature runs on a new intelligence layer that responds to feedback, and Netflix calls it a "test kitchen" for new discovery features. It's rolling out to a small group first, with wider access coming soon.

⚙️ DeepSeek open-sources self-managing Super-agent ⚙️

DeepSeek has published Super-agent, a Router/Worker/Watchdog architecture built for speed when running many agent sessions without human supervision. In simulated tests it nearly halved 95th-percentile task time, and the dsh-superagent prototype is open source on GitHub, adding competitive pressure on Western agent frameworks. The design splits work across specialized roles so long-running agents can operate autonomously at scale, with less human babysitting.