🎉 Open-source milestone: Our community has now uploaded over 1 million models to the Hub! From tiny edge models to massive 405B parameter behemoths, the open-source AI ecosystem is thriving. Thank you to every contributor! 🤗 #OpenSource #AI

Stable Diffusion 4.0 is here. Key improvements: • 16K resolution support • Real-time video generation at 60fps • Architectural understanding for interior design • 3D model generation from single images Free for personal use. Commercial licenses available.

Hot take: Current LLMs are NOT intelligent. They are sophisticated pattern matchers trained on internet text. True AI will require a fundamentally different architecture — one that can reason, plan, and learn from limited data like humans do.

Excited to announce AlphaFold 3 — our latest breakthrough in protein structure prediction. It can now predict the structure of nearly all molecular interactions in biology, including DNA, RNA, and small molecules. This could revolutionize drug discovery. 🧬

Introducing Gemini Ultra 2.0 — our most capable model yet. It outperforms GPT-5 on 7 out of 10 benchmarks and introduces native multimodal reasoning across text, images, audio, and video. Available today in Google Cloud and Bard.

AI is the new electricity. Just as electricity transformed every major industry 100 years ago, AI is now poised to do the same. The key is not just building AI, but deploying it responsibly across healthcare, education, and manufacturing.

Just finished a deep dive into the Transformer architecture internals. The attention mechanism is still one of the most elegant ideas in ML. Here's my new video tutorial breaking it down from scratch with code: [link] Thread 🧵👇

We're releasing GPT-5 today. It achieves state-of-the-art performance on every major benchmark, with dramatically improved reasoning, coding, and multimodal capabilities. Try it now at chat.openai.com 🚀