*Significant progress in AI and Robotics this week.* 🚀 So, I summarized everything from Nvidia, OpenAI, ByteDance, Google, Figure, Mistral, Borg, Apple, Meta, and more. Here's everything you need to know and how to make sense out of it: * *OpenAI 🤖 added Deep Research in ChatGPT to do extensive web analysis* on complex topics and deliver research reports under 30 min. It uses a specialized version of o3 to analyze text, images, and PDFs across sources. Only available to Pro users right now. * *ByteDance demoed OmniHuman-1, an AI that generates deepfake videos* from a single image & audio input. It's trained on 19k hrs of video, AI can handle diverse inputs while maintaining style-specific motion. It's not publicly available, but still wild. * *Google unveiled new Gemini 2.0 models* - Pro Experimental & Flash Lite - while making the original Flash GA. Pro Exp. with 2M token context window excels at coding and complex prompts, while Flash Lite outperforms 1.5 Flash at the same speed, price * *Amid DeepSeek hype, French AI lab Mistral launched its ‘le Chat’* assistant on iOS and Android. The app offers web search, doc processing, code interpreter, and image generation with the ability to deliver flash answers at 10x the speed of ChatGPT, Claude. * *Meta AI researchers introduced VideoJam,* an AI framework for generating videos with realistic motion, dynamics and physics. It can be used to fine-tune any text-to-video model, significantly improving its motion quality, without any extra data or scaling.