AI workflow automation, LLMs leaderboard rankings, Skybox images from a sketch, Windows copilot and more
Greetings and welcome to this week's AI Brews - your thoughtfully curated guide to AI products, learning resources and a concise roundup of the week's impactful news. Our goal? To provide a balanced selection in the rapidly evolving AI landscape, keeping you well-informed without the information overload. We value your feedback - don't hesitate to reply to this email with suggestions on how we can make this better for you. Thanks!
In today’s issue:
AI Pulse: News, Insights and Social Spotlight of the Week
AI Toolbox: Product Picks of the Week
AI Skillset: Learn & Build
🗞️ AI Pulse: News, Insights and Social Spotlight of the Week
🔥 News & Insights
Meta released a new open-source model, Massively Multilingual Speech (MMS) that can do both speech-to-text and text-to-speech in 1,107 languages and can also recognize 4,000+ spoken languages. Existing speech recognition models only cover approximately 100 languages out of the 7,000+ known spoken languages. [Details | Research Paper | GitHub].
New research presented in the paper ‘QLORA: Efficient Finetuning of Quantized LLMs’ makes it possible to train and fine-tune LLMs on consumers' GPUs. Their new open-source model Guanaco, outperforms all previous openly released models on the Vicuna benchmark, reaching 99.3% of the performance level of ChatGPT while only requiring 24 hours of finetuning on a single GPU [Paper | GitHub | Huggingface].
Adobe has integrated its generative AI model Firefly, into the Photoshop desktop app via a new tool, Generative Fill. Users can use natural language prompts to create and do complex image edits in Photoshop [details].
Jugalbandi, a chatbot developed in collaboration between Microsoft, OpenNyAI, AI4Bharat and Indian government, provides rural Indians with information on government schemes in 10 local languages via WhatsApp, overcoming language barriers [Details].
Google’s AI-based flood forecasting platform 'Flood Hub' is now available in 80 countries, offering predictions up to a week in advance [Details].
Microsoft’s AI centric announcements at Build 2023 conference:
Windows Copilot - Centralized AI assistance in Windows 11, accessible from the taskbar across all applications. Users can ask copilot to customize settings, perform tasks ranging from simple on-screen text summarization to complex ones requiring multiple app interactions. Bing Chat plugins will be available in Windows Copilot[Details | Youtube Link].
Microsoft has adopted OpenAI's open plugin standard for ChatGPT. This will enable developers to build plugins once that work across ChatGPT, Bing, Dynamics 365 Copilot and Microsoft 365 Copilot [Details].
Launch of copilot in Power Pages, Microsoft’s low-code tool for creating data-centric business websites. The AI Copilot will enable users to generate text, build detailed forms and chatbots as well as help in page creation, site theming & image generation via text prompts [Details].
Azure AI Studio: users can build a custom chat assistant based on OpenAI’s models trained on their own data .
Microsoft Fabric: a new end-to-end data and analytics platform.that will include copilot for users to build data pipelines, generate code, build machine learning models and more [Details].
AI generated images by Bing Image Creator and Microsoft Designer will have origin clearly disclosed in the image’s metadata [Details].
Meta announced a new language model LIMA (Less Is More for Alignment), based on 65B LLaMa that achieves comparable or better responses than GPT-4 and Bard by fine-tuning only on 1k supervised samples [Details].
Skybox AI, the free 360° image generator tool by Blockade labs, now supports creating a skybox from a sketch, generation & downloading of depth maps (on desktops and tablets) as well as negative prompting [Link].
See the latest leaderboard rankings for large language models (LLMs) by Chatbot Arena - a benchmark platform for LLMs, by LMSYS Org, that features anonymous, randomized battles in a crowdsourced manner [Details].
Intel plans to create a series of generative AI models, with 1 trillion parameters, for the scientific research community [Details].
BLOOMChat, a new, open, 176 billion parameter multilingual chat LLM, built on top of BLOOM has been released by SambaNova and Together and is available for commercial use. BLOOM is already the largest multilingual open model, trained on 46 languages and developed by an international collaboration of more than 1000 researchers [Details]..
OpenAI is is launching a program to award ten $100,000 grants to fund experiments in setting up a democratic process for deciding what rules AI systems should follow [Details].
Google announced Product Studio - a new tool for merchants to create product images using generative AI [Details].
Character.AI, the popular AI-powered web app that lets users create and chat with their favourite characters, has launched mobile Apps for iOS and Android [Details].
Google DeepMind's visual language model, Flamingo, is improving video search results by generating descriptions for YouTube Shorts. Also, their AI model, MuZero, is optimizing video compression for YouTube's live traffic [Details].
ChatGPT updates: a. Shared Links that will enable users to share favourite ChatGPT conversations through a unique URL, allowing others to see and continue the dialogue. b. Bing is the default search engine for ChatGPT and this will soon be accessible to all free ChatGPT users via a plugin [Details].
OpenAI predicts that ‘within the next ten years, AI systems will exceed expert skill level in most domains, and carry out as much productive activity as one of today’s largest corporations’ and suggests an international regularity authority [Details: ‘Governance of superintelligence’].
🔦 Social Spotlight
A new social media app, Airchat by Naval Ravikant [Link with demo ].
Agent Weekend - Workshop & Hackathon Co-hosted by Codium AI & AutoGPT. Founder AutoGPT shares the roadmap [Youtube video].
DragGAN integrated into InternGPT - an open source demo platform where you can easily showcase your AI models [Link]
Wharton School's Prof. Ethan Mollick asks students to use Bing for assignment: Formulate 'Impossibly Ambitious' business Ideas and simulate critique from famous founders [Link]
Building an end to end product prototype using AI and Replit in 2 days for a hackathon [Link].
🔍 🛠️ AI Toolbox: Product Picks of the Week
Levity
Levity is a powerful AI-based tool that automates everyday tasks without coding. It lets you create custom AI models trained on your own data for classification of images, documents or text and integrate it with other tools in your workflow. All this can be done visually without any technical knowledge or writing any code. Levity can, for example, manage a Gmail inbox by smartly sorting emails or analyze product images for appropriate categorization. Check out the success stories on their website which lists some interesting use cases such as microscopic image processing for worm egg counting, or real estate image classification to assess property features and their impact on bookings.
pixian.ai
An AI-powered background image removal tool that is free while in beta and doesn’t require any login.
Dante
A GPT-4 powered custom no-code chatbot builder for your website, trained on your data. Supports multiple file types, website links as well as images and videos.
📕 AI Skillset: Learn & Build
A list of open LLMs available for commercial use [GitHub Link].
A big curated list of AI learning resources, AI Canon, by a16z
Short video tutorials by Adobe on the new Generative Fill feature in Photoshop:
Learn how to fine-tune large language models (LLMs) on a custom dataset by using Lit-Parrot, a nanoGPT based implementation of the GPT-NeoX
model that supports – StableLM, Pythia, and RedPajama-INCITE model weights. [Link].Camel + LangChain for Synthetic Data & Market Research:
Thank you for reading AI Brews! If you have any thoughts, questions or just want to say hello, please don't hesitate to hit reply. Mariam