[00:00:00] Intro and welcome everyone[00:00:52] GPT4 - Vision from OpenAI[00:05:06] Safety concern with GPT4-V[00:09:18] GPT4 can talk and listen as well[00:12:15] Apple rumors, on device inference, and Siri[00:17:01] OpenAI Voice Cloning Tech used in Spotify to translate podcasts[00:19:44] On the risks of Voice Cloning tech being open sourced[00:26:07] Alex statement on purpose of ThursdAI[00:27:53] “AGI has been achieved internally”;[00:32:10] OpenAI, Jonny Ive and Masa are rumored to be working on a hardware device[00:33:51] Cloudflare AI - Serverless GPU on global scale[00:37:13] Cloudflare AI partnership with HuggingFace to allow you to run many models in your own[00:40:34] Cloudflare announced the Vectorize DB and embedings on edge[00:46:52] Cloudflare AI gateway - proxy LLM calls, caching, monitoring, statistics and fallback[00:51:15] Part 2 - intro an recap[00:54:14] Meta AI announcements, bringing AI agents to 3 billion people next month[00:56:22] Meta announces EMU image model to be integrated into AI agent on every platform[00:59:38] Meta RayBan glasses upgraded to spatial computing, with AI and camera access[01:00:39] On the topic os smart glasses, GoogleGlass, and the acceptance society wide to have[01:05:37] Safety and societal implications of everyone having glasses and recording everything[01:12:05] Part 3 - Open Source LLMs, Mistral, QWEN and CapyBara[01:21:27] Mistral 7B - SOTA 7B general model from MIstralAI[01:23:08] On the topic of releasing datasets publically and legal challenges with obtaining that[01:24:42] Mistral GOAT team giving us a torrrent link to a model with Apache 2 license.Truly, as I’ve been doing these coverages in one form or another for the past 9 months, and I don’t remember a week this full of updates, news, state of the art open source models and more.So, here’s to acceleration (and me finally facing the fact that I need a niche, and decide what I’ll update on and what I won’t, and also be transparent with all of you about it)On a separate note, this past two weeks, ThursdAI had exposure to Yann Lecun (RTs), joined on stage by VP of DevRel in Cloudflare and their counterpart in HuggingFace, CEO of Anaconda joined us on stage this episode and we’ve had the chief scientist of Mistral join in the audience ? ThursdAI really shapes to be the place where this community meets, and I couldn’t be more humbled and prouder of the show, the experts on stage that join from week to week, and the growing audience ?‍♂️ ok now let’s get to the actual news!ThursdAI - Weeks like this one highlight how important it is to stay up to date on many AI news, subscribe, I’ve got some cool stuff coming! ?All right so here’s everything we’ve covered on ThursdAI, September 28th:(and if you’d like to watch the episode video with the full transcript, it’s here for free):Show Notes + Links* Vision* ? Open AI announces GPT4-Vision (Announcement, Model Card)* Meta glasses will be multimodal + AI assistant (Announcement)* Big Co + API updates* Cloudflare AI on workers, serverless GPU, Vector DB and AI monitoring (Announcement, Documentation)* Cloudflare announces partnerships with HuggingFace, Meta* Claude announces $4 billion investment from Amazon (Announcement)* Meta announces AI assistant across WhatsApp, Instagram* Open Source LLM* ? Mistral AI releases - Mistral 7B - beating LLaMa2 13B (Announcement, Model)* Alibaba releases Qwen 14B - beating LLaMa2 34B (Paper, Model, Vision Chat)* AI Art & Diffusion* Meta shows off EMU - new image model* Still waiting for DALL-E3 ?* Tools* Spotify translation using Open AI voice cloning techVisionGPT 4-VisionI’ve been waiting for this release since March 14th (literally) and have been waiting and talking about this on literally every ThursdAI, and have been comparing every open source multimodality image model (IDEFICS, LlaVa, QWEN-VL, NeeVa and many others) to it, and none came close!And here we are, a brief rumor about the upcoming Gemini release (potentially a multimodal big