Meta announced new prescription AI glasses, which might sound like just a hardware update, but it’s actually more interesting because of the software features they are adding. The glasses are designed so people can wear them all day as normal glasses, not just as a gadget. If AI glasses are going to become mainstream, they probably need to replace something people already wear every day. The new AI features they are adding include things like hands-free nutrition tracking, WhatsApp message summaries, recalling information from chats, writing messages by drawing with your finger on a surface, navigation in the lens, widgets like weather and calendar, and display recording from the glasses. We are slowly getting closer to a real Jarvis an AI that is always there, can see what you see, hear what you hear, and help you throughout the day. Meta is basically betting that AI glasses are the future interface for AI.

OpenAI just announced they raised $122 billion at a $852 billion valuation. At this point they are basically building AI infrastructure, not just models. Some numbers from their announcement: • 900M weekly ChatGPT users • 50M subscribers • About $2B revenue per month • Enterprise already 40% of revenue • 15B tokens per minute through their API Their strategy is basically a flywheel: more compute leads to better models, better models lead to better products, better products lead to more users and revenue, and that revenue gets reinvested into more compute and infrastructure. They are also moving toward what they call an AI superapp that combines ChatGPT, coding, browsing, agents, and workflows into one system. https://openai.com/index/accelerating-the-next-phase-ai/

Would anyone here be interested in poking around my app and giving me some feedback? I have had some sold beta testers but would love some more now that I've made a ton of updates.

StudyPilot — AI-powered exam preparation with adaptive learning The problem: 70% of students abandon education apps due to poor usability. Generic learning apps deliver one-size-fits-all content with no adaptation to individual knowledge gaps. Students preparing for professional exams (ACCA, CFA, SQE, USMLE) or university exams need targeted, gap-aware study tools — not just flashcard apps. 80% of learners prefer personal devices, and mobile learning engagement increases up to 40% with gamification. The app: Students select their exam or course. An NLP engine processes study materials (textbook photos taken with a camera, PDF uploads, pasted notes) and generates question banks, flashcards, and practice scenarios. ML models track performance across topics, identify knowledge gaps, and dynamically adjust the study plan — spending more time on weak areas. An AI tutor answers questions conversationally via text or voice. ML-optimised spaced repetition scheduling delivers push notifications at scientifically optimal review times. Any thoughts on this app idea?

What happened in the last 24 hours: A new model called Qwen 3.6 Plus Preview was released on OpenRouter and it’s currently free to use. Qwen models have been improving very quickly, especially for coding and reasoning tasks, and they are often much cheaper than the big frontier models. Many people are starting to use Qwen models for agents and coding workflows because of the price and performance. If you want to try it out, this is a good time while it’s free. It works with tools like Cursor, OpenCode, Kilo Code, and anything that supports OpenRouter. Just note that the free versions specifically are known to use prompts and conversations for training, so be careful with sensitive code or data. https://openrouter.ai/qwen/qwen3.6-plus-preview:free Another interesting thing: Claude Code’s source code reportedly leaked through a map file that was published in their npm registry. The repository is already on GitHub if you want to look through it: https://github.com/instructkr/claude-code It’s actually a rare chance to look under the hood and see how a tool like Claude Code is built and structured. Ollama also announced MLX support, Apple’s machine learning framework for Apple Silicon. This means running local models on Apple Silicon machines like MacBooks and Mac Minis, and setups like OpenClaw, should become faster and more efficient. Local AI on personal machines is improving very quickly, and it’s becoming more realistic to run useful local models locally instead of always using servers. https://ollama.com/blog/mlx