New open and commercially available models, Emu image model by Meta, voice and image capabilities in ChatGPT , open-source multiple agents framework by Microsoft, and more
Greetings and welcome to this week's AI Brews for a concise roundup of the week's major developments in AI.
In today’s issue (Issue #33):
AI Pulse: Weekly News & Insights at a Glance
AI Toolbox: Product Picks of the Week
AI Skillset: Learn & Build
🗞️🗞️ AI Pulse: Weekly News & Insights at a Glance
🔥 News
Meta AI presents Emu, a quality-tuned latent diffusion model for generating highly aesthetic images. Emu significantly outperforms SDXLv1.0 on visual appeal [Paper].
Meta AI researchers present a series of long-context LLMs with context windows of up to 32,768 tokens. LLAMA 2 70B variant surpasses gpt-3.5-turbo-16k’s overall performance on a suite of long-context tasks [Paper].
Abacus AI released a larger 70B version of Giraffe. Giraffe is a family of models that are finetuned from base Llama 2 and have a larger context length of 32K tokens [Details].
Meta announced [Details]:
Meta AI - a new AI assistant users can interact with on WhatsApp, Messenger and Instagram. Will also be available on Ray-Ban Meta smart glasses and Quest 3, Meta’s mixed reality headset.
AI stickers that enable users to generate customized stickers for chats and stories using text. Powered by Llama 2 and the new foundational model for image generation, Emu.
28 AI characters, each with a unique personality that users can message on WhatsApp, Messenger, and Instagram.
New AI editing tools, restyle and backdrop in Instagram.
AI Studio - a platform that supports the creation of custom AIs by coders and non-coders alike.
Cerebras and Opentensor released Bittensor Language Model, ‘BTLM-3B-8K’, a new 3 billion parameter open-source language model with an 8k context length trained on 627B tokens of SlimPajama. It outperforms models trained on hundreds of billions more tokens and achieves comparable performance to open 7B parameter models. The model needs only 3GB of memory with 4-bit precision and takes 2.5x less inference compute than 7B models and is available with an Apache 2.0 license for commercial use [Details].
OpenAI is rolling out, over the next two weeks, new voice and image capabilities in ChatGPT enabling ChatGPT to understand images, understand speech and speak. The new voice capability is powered by a new text-to-speech model, capable of generating human-like audio from just text and a few seconds of sample speech. [Details].
Mistral AI, a French startup, released its first 7B-parameter model, Mistral 7B, which outperforms all currently available open models up to 13B parameters on all standard English and code benchmarks. Mistral 7B is released in Apache 2.0, making it usable without restrictions anywhere [Details].
OpenAI has returned the ChatGPT browsing feature for Plus subscribers, enabling ChatGPT to access internet for current information. It was disabled earlier as users were able to deploy it to bypass the paywalls of leading news publishers [Details].
Microsoft has released AutoGen - an open-source framework that enables development of LLM applications using multiple agents that can converse with each other to solve a task. Agents can operate in various modes that employ combinations of LLMs, human inputs and tools [Details].
LAION released LeoLM, the first open and commercially available German foundation language model built on Llama-2 [Details]
Researchers from Google and Cornell University present and release code for DynIBaR (Neural Dynamic Image-Based Rendering) - a novel approach that generates photorealistic renderings from complex, dynamic videos taken with mobile device cameras, overcoming fundamental limitations of prior methods and enabling new video effects [Details].
Cloudflare launched Workers AI (an AI inference as a service platform), Vectorize (a vector Database) and AI Gateway with tools to cache, rate limit and observe AI deployments. Llama2 is available on Workers AI [Details].
Amazon announced the general availability of Bedrock, its service that offers a choice of generative AI models from Amazon itself and third-party partners through an API [Details].
Google announced it’s giving website publishers a way to opt out of having their data used to train the company’s AI models while remaining accessible through Google Search [Details].
Spotify has launched a pilot program for AI-powered voice translations of podcasts in other languages - in the podcaster’s voic. It uses OpenAI’s newly released voice generation model [Details].
Getty Images has launched a generative AI image tool, ‘Generative AI by Getty Images’, that is ‘commercially‑safe’. It’s powered by Nvidia Picasso, a custom model trained exclusively using Getty’s images library [Details].
Optimus, Tesla’s humanoid robot, can now sort objects autonomously and do yoga. Its neural network is trained fully end-to-end [Link].
Amazon will invest up to $4 billion in Anthropic. Developers and engineers will be able to build on top of Anthropic’s models via Amazon Bedrock [Details].
Google Search indexed shared Bard conversational links into its search results pages. Google says it is working on a fix [Details].
Pika Labs' text-to-video tool now lets users encrypt a message in a video [Twitter Link].
🔦 Weekly Spotlight
How AI-powered echoes are making waves in the fight against heart failure [Link].
AI language models can exceed PNG and FLAC in lossless compression, says study [Link].
Everyone is above average. Is AI a Leveler, King Maker, or Escalator? [Link].
What Builders Talk About When They Talk About AI [Link].
The Llama Ecosystem: Past, Present, and Future [Link].
🔍 🛠️ AI Toolbox: Product Picks of the Week
Animant: Create a 3D model out of anything from a physical object to an entire floor plan, and animate the movement of objects in your space by writing words
Sizzle: An AI-powered learning app that breaks down any problem( whether it's math or chemistry, multiple choice or word problems) into easy-to-follow steps .
Durable: AI website builder that generates an entire website with images and copy.
📕 📚 AI Skillset: Learn & Build
Creating on-brand backdrops with Midjourney: Ingredients Edition [Link].
Getting to know Llama 2: Everything you need to start building - CoLab notebook from workshop at Meta Connect 2023 [Link].
Writing poems using LLama 2 on Workers AI [Link].
Pair Programming with a Large Language Model - New short course (free) on DeepLearning.AI in collaboration with Google [Link].
AI Brews is free, and your sharing it with a friend helps us grow. Thanks for your support and have a nice weekend! 🎉 Mariam.