Keep Up with the AI Times
All the latest developments in AI that you just can't afford to miss!
Hey there folks 👋
Firstly thanks a lot for the overwhelming response and support you have given to the publication. This is what will keep me going further.
Welcome to the second issue from Unwind AI, your one-stop shop for all things trending in the AI and NLP space. This issue covers:
Latest Developments 🌍
News from the Industry 🧑🏫
Tools of the Trade ⚒️
Hot Takes 🔥
AI Meme of the Week 🤡
Let’s straight cut to the chase!
Latest Developments 🌍
PyTorch 2.0 🐍
PyTorch 2.0 got released with the most awaited feature "torch.compile” that will push its performance to new heights. It offers the same user experience and eager-mode development as previous versions but introduces new technologies that improve performance and support for dynamic shapes and distributed computing.
PyTorch 2.0 is fully backward compatible and can provide speedups of up to 43% on server-class GPUs. It is currently in preview, with plans for a stable release in early 2023.
Stable Diffusion v2 🖼️
Stability AI has recently announced the release of its text-to-image model Stable Diffusion 2.0 with ground-breaking enhancements to the earlier version. Can’t wait to discuss the improvements!
Text encoder: The 2.0 version uses OpenCLIP-ViT/H as the text encoder, developed by LAION, that has shown to achieve 78% zero-shot accuracy and bagged the title of the best open-source CLIP model as of September 2022.
Training data: The successor version is trained on an aesthetic subset of LAION-5B which is a dataset of 5.85 billion CLIP-filtered image-text pairs from 100+ other languages besides English, and further filtered for adult content.
Super-resolution Upscaler Diffusion Model: The new version also includes an Upscaler Diffusion model that enhances the resolution of the low-resolution generated images.
(Left) 128x128p resolution image upscaled to (Right) 512x512p image by Upscaler Diffusion Model
Depth-to-Image Diffusion Model: The new depth-guided model called the depth2img is the coolest of all! The model understands the depth of the image and based on the prompt and the depth, it can generate more images while restoring the nuances of the original image.
Updated Inpainting Diffusion Model: Lastly, the 2.0 version includes a text-based inpainting model allowing the user to switch or fill parts of the input image in a jiffy.
Character AI - An AI with Personality 💁🏻♂️
Introducing Character AI, the world's first platform for creating your own personal AI assistant. With just a few clicks, you can design and train your own conversational AI agent, capable of handling dynamic tasks and exchanging ideas with you. Imagine having your own personal teacher, assistant, or even friend, always available to help you with anything you need.
Character AI’s beta is already generating 1 billion words per day and hosting more than 350 thousand unique characters. Some of the most interesting usecases are chatting with Life Coach, Socrates, and Elon Musk!
Blueprint by Baseten 🕹️
Baseten recently announced the launch of Blueprint, bringing the most cutting-edge generative AI models like Stable Diffusion, Whisper, and Flan-T5 to the builders locally, allowing them to fine-tune these prebuilt models into their APIs. Not just this, Blueprint also provides a comprehensive web IDE to turn fine-tuned models into a hosted API. The IDE includes features like:
A Python console with developer experience upgrades for working with model inputs and outputs
Built-in testing tools and logs with API routes as Python functions
Automatic deployments and fully managed infrastructure
Spellbook by Scale AI 🪄
Scale AI has recently launched Spellbook, a platform that makes it easy to take advantage of powerful LLMs. It comes up with the following features:
Create: Upload your data and easily review and edit your prompts.
Compare: Quickly compare experiments across different LLMs and prompts.
Finetune: Finetune on your existing data to continuously improve model performance.
Deploy: Deploy promising variants to production-ready API endpoints with built-in monitoring and analytics in one click.
🕹️ Try it out yourself, now available on early access!
News from the Industry 🧑🏫
First-ever AI Film Festival 🍿
Runway, a platform for creating graphics content using AI, is hosting the first-ever annual AI Film Festival. There will be a panel of renowned filmmakers, creators, and AI innovators to review all films, and finalists will be showcased at a screening in New York. Additionally, participants will be eligible for $10k+ in prizes and a complimentary two-year Runway subscription. Submissions close on January 15th, 2023!
OpenAI’s Whisper v2 🎶
OpenAI quietly dropped Whisper v2 this week while people were still in the aww of playing with ChatGPT. Whisper v2 works on the same architecture as the original old model but has improved performance. The "large-v2" model is trained for more epochs with regularization which makes it able to perform better than v1.
🕹️ Try it out yourself - OpenAI Whisper
Evaluating GPT-3.5 🤔
Check out this blog post by Scale AI, taking a closer look at the release of the new model “text-davinci-003” to GPT-3. It examines the language model's capabilities and its potential applications, from creative writing to text summarization, in comparison to its predecessor 002.
ChatGPT Explained 📖
ChatGPT is by far the coolest AI that has taken the internet by storm. Check out this wonderful explanation of ChatGPT by Yannic Kilcher where he will take you on this amazing journey of unbelievable AI progress 👇
Tools of the Trade ⚒️
AI Toolbot - Build AI tools on Autopilot 🤫
Introducing AI toolbot by Zahid Khawaja that uses GPT-3 to spin off new GPT-3 tools on-the-fly. Now building AI tools is as simple as asking the AI toolbot in plain English. Don’t believe me, try it yourself!
Rick and Mortify 🧪
Creative space being occupied by a handful of people is a thing of the past now when creative AI is leading the way to generate superior quality, long-form content. A profound example is Rick and Mortify by Julia Turc and Mihail Eric.
Inspired by the timeless sitcom Rick and Morty, this platform lets you create your own version of stories of the eccentric scientist Rick and his grandson Morty, using GPT-3 and DALL.E 2. Check out this storyboard created using the prompt “Rick and Morty go on a trip to the lost city of Atlantis”, for instance.
Hume - your AI hiring assistant 🧑🏭
Hume is an AI-powered tool that makes your hiring process more efficient by:
transcribing,
summarizing, and
Q&A extraction
By extracting questions and categorization, you’d know what competencies and skills you have discussed and whether the candidate matches those required for the job. This will bring evidence to your hiring decisions as well as reduce the time per hire.
Hot Takes 🔥
First Prompt Engineer Ever 😱
ChatGPT to the moon 🚀
An AI assistant or a full-time Author 📕
AI meme of the week 🤡
That’s all for this week folks!
Will see you guys next Saturday with more such content. Don’t forget to subscribe and provide your feedback below!
BONUS 🎉
Share this newsletter with three other friends and stand a chance to win a signed copy of my book Neural Search - From Prototype to Production with Jina. Winners will be selected on a monthly basis!
🎁 Every paid subscriber will also receive $39 USD worth of learning resources on trending topics like Python, Data Science, Machine Learning, and NLP!