ThursdAI November 2ndHey everyone, welcome to yet another exciting ThursdAI. This week we have a special announcement, the co-host of and I will be hosting a shared X space live from Open AI Dev Day! Monday next week (and then will likely follow up with interviews, analysis and potentially a shared episode!)Make sure you set a reminder on X (https://thursdai.news/next) , we’re going to open the live stream early, 8:30am on Monday, and we’ll live stream all throughout the keynote! It’ll be super fun!Back to our regular schedule, we covered a LOT of stuff today, and again, were lucky enough to have BREAKING NEWS and the authors of said breaking news (VB from HuggingFace and Emozilla from Yarn-Mistral-128K) to join us and talk a little bit in depth about their updates![00:00:34] Recap of Previous Week's Topics[00:00:50] Discussion on AI Embeddings[00:01:49] Gradio Interface and its Applications[00:02:56] Gradio UI Hosting and its Advantages[00:04:50] Introduction of Baklava Model[00:05:11] Zenova's Input on Distilled Whisper[00:10:32] AI Regulation Week Discussion[00:24:14] ChatGPT new All Tools mode (aka MMIO)[00:35:45] Discussion on Multimodal Input and Output Models[00:36:55] BREAKING NEWS: Mistral YaRN 7B - 128K context window[00:37:02] Announcement of Mistral Yarn Release[00:46:47] Exploring the Limitations of Current AI Models[00:47:25] The Potential of Vicuna 16k and Memory Usage[00:49:43] The Impact of Apple's New Silicon on AI Models[00:51:23] Introduction to New Models from Nius Research[00:51:39] The Future of Long Context Inference[00:53:42] Exploring the Capabilities of Obsidian[00:54:29] The Future of Multimodality in AI[00:58:48] The Exciting Developments in CodeFusion[01:06:49] The Release of the Red Pajama V2 Dataset[01:12:07] The Introduction of Luma's Genie[01:16:37] Discussion on 3D Models and Stable Diffusion[01:17:08] Excitement about AI Art and Diffusion Models[01:17:48] Regulation of AI and OpenAI Developments[01:18:24] Guest Introduction: VB from Hug& Face[01:18:53] VB's Presentation on Distilled Whisper[01:21:54] Discussion on Distillation Concept[01:27:35] Insanely Fast Whisper Framework[01:32:32] Conclusion and RecapShow notes and links:* AI Regulation* Biden Executive Order on AI was signed (Full EO, Deep dive)* UK AI regulation forum (King AI speech, no really, Arthur from Mistral)* Mozilla - Joint statement on AI and openness (Sign the letter)* Open Source LLMs* Together AI releases RedPajama 2, 25x larger dataset (30T tokens) (Blog, X, HF)* Alignment Lab - OpenChat-3.5 a chatGPT beating open source model (HF)* Emozilla + Nous Research - Yarn-Mistral-7b-128k (and 64K) longest context window (Announcement, HF)* LDJ + Nous Research release Capybara 3B & 7B (Announcement, HF)* LDJ - Obsidian 3B - the smallest open source multi modal model (HF, Quantized)* Big CO LLMs + APIs* ChatGPT "all tools" MMIO mode - Combines vision, browsing, ADA and DALL-E into 1 model (Thread, Examples, System prompt)* Microsoft CodeFusion paper - a tiny (75M parameters) model beats a 20B GPT-3.5-turbo (Thread, ArXiv)* Voice* Hugging Face - Distill whisper - 2x smaller english only version of Whisper (X, paper, code)* AI Art & Diffusion & 3D* Luma - text-to-3D Genie bot (Announcement, Try it)* Stable 3D & Sky changerAI Regulation IS HERELook, to be very frank, I want to focus ThursdAI on all the news that we're getting from week to week, and to bring a positive outlook, so politics, doomerism, and regulation weren't on the roadmap, however, with weeks like these, it's really hard to ignore, so let's talk about this.President Biden signed an Executive Order, citing the old, wartime era Defence Production act (looks like the US gov. also has "one weird trick" to make the gov move faster) and it wasn't as bombastic as people thought. X being X, there has been so many takes pre this executive order even releasing about regulatory capture being done by the big AI labs, about how open source is no longer going to be possible, and if