AI Advancements, At Your Fingertips!
Our weekly AI report for the curious minds out there!
Hey there 👋
Welcome to another exciting edition of Unwind AI with our weekly scoop on the most innovative and revolutionary developments in the field of AI. With the rapid advancement of AI, there is always something new to learn and explore, and we are thrilled to be your guide in navigating the world of cutting-edge technology. We'll be taking a closer look at some of the breakthroughs in natural language processing, emerging career trends, companies leading the race in the generative AI field, and more juice details!
So, sit back, relax, and get ready to dive into the deep pool of AI!
This issue covers:
Latest Developments 🌍
News from the Industry 🧑🏫
Tools of the Trade ⚒️
Hot Takes 🔥
AI Meme of the Week 🤡
Latest Developments 🌍
Effortless Editing Experience ✂️
CMU and Adobe’s paper proposes pix2pix-zero, an image-to-image translation method that allows for on-the-fly editing of real and synthetic images using pre-trained text-to-image diffusion models like Stable Diffusion, without requiring manual text prompting or costly fine-tuning. The method automatically discovers editing directions and uses cross-attention guidance to preserve the content structure of the original image.
The Multi-Speaker Speech Wizard 🎤
Google Research has proposed SPEAR-TTS, a text-to-speech model that can convert written text into spoken words with natural-sounding voices, even for multiple speakers, with very little supervision needed. It breaks the process down into two parts: understanding the text and converting it into a set of instructions for speaking. This allows the system to learn from audio recordings of people speaking, rather than needing a lot of text-to-speech examples.
How to Design Smarter, Not Harder 🪄
Diagram has introduced a preview of its new AI design assistant called Genius, which works within the popular design tool Figma. Genius can understand what you are designing and suggest components from your design system to autocomplete your design. The ultimate goal is to generate user interfaces and allow everyone to design their digital world.
Teaching LLMs to Teach Themselves 👩🏫
Microsoft Research has introduced a method called Synthetic Prompting that helps large language models perform reasoning tasks through step-by-step demonstrations. The method leverages a few handcrafted examples to prompt the model to generate more examples by itself and selects effective demonstrations to elicit better reasoning. The method alternates between a backward and forward process to generate new examples.
A Giant Leap in Healthcare with AI 🦘
The MedARC (Medical AI Research Center) has been launched in partnership with Stability AI, with a novel, open and collaborative approach to advancing the field of AI in healthcare. The progress of AI in other fields has been significant but has not had relevant use cases in medical AI as these models are not trained with medical data specifically.
From Good to Great through Hindsight Finetuning 👌
Berkeley has proposed a novel technique called Hindsight Finetuning to address the limitation of supervised finetuning in learning from both positive and negative human feedback. The approach is inspired by how humans learn from hindsight experience, and involves conditioning the model on a sequence of model generations paired with hindsight feedback and finetuning the model to predict the preferred output.
News from the Industry 🧑🏫
The Dark Side of AI-Powered Search Engines 🌑
The integration of large language models into search engines by tech giants like Google, Microsoft, and Baidu may significantly enhance search capabilities, but at the cost of environment. It will lead to a dramatic rise in energy consumption and carbon emissions. GPT-3 training alone consumed 1,287 MWh and led to over 550 tons of CO2 emissions. If left unchecked, the rise of generative AI in search could worsen the environmental impact of cloud computing.
(Source: Wired.com)
OpenAI Joins Office Suite 🏢
Microsoft to showcase OpenAI in core apps like Word, PowerPoint & Outlook, improving search, email replies & document writing using ChatGPT language technology. Driven by ChatGPT's 100 million monthly users, Microsoft plans to harness AI's potential in 2023.
(Source: The Verge)
On-Brand Content Creation in a Flash ⚡
Jasper for Business is here! It aims to help businesses create high-quality content faster by reducing the time spent on first drafts by up to 80%, resulting in a 10x output of high-quality content and potentially leading to a 400% ROI. The platform can be tailored to a business's unique brand voice and style guide, allowing it to consistently generate on-brand content across a team. Jasper can also be added as an extension to Chrome and directly be used for Gmail and Google Docs.
![[video-to-gif output image]](https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F025b93ef-a5b3-471d-beb5-38416581b887_800x450.gif "[video-to-gif output image]")
Opera Joins the Conversational Search Engine Race 🏇
Opera is adding AI-generated content services to its browsers, starting with a 'Shorten' button for AI-generated summaries of webpages and articles. This marks the company's expansion of its AI program, aiming to provide users with a filtered and relevant summary of online content. AI is the key to an AI-powered web, according to Opera.
Guess the Prompt 🤑
Stable Diffusion is hosting a competition that challenges participants to create a model that can predict the text prompt that generated a given image. The goal is to understand how reversible the latent relationship between the image and prompt is. The competition requires participants to submit embeddings of their predicted prompts to calculate prompt similarity in a robust way. Winners will be offered prizes of up to $12,000! The submissions are starting from February 13, 2023.
Cohere flying under the Radar 🛸
Cohere AI has been flying under the radar, according to its co-founder and CEO Aidan Gomez, despite offering foundation models that are competitive with those of industry giants such as Google and OpenAI. However, this may be about to change, as Reuters reports that Cohere is in talks to raise hundreds of millions of dollars in a funding round that could value the startup at more than $6 billion, which highlights the growing investment in generative AI.
(Source: Venture Beat)
Bing AI's Demo Disaster 💥
While people are going gaga over Microsoft’s new Bing search engine, it made mistakes during its demo which largely went unnoticed and did not receive the same criticism that Google received when its Bard AI provided an incorrect answer during an ad. Here’s a detailed analysis of the factual errors made by Bing.
While there’s way long to go, let’s see who will come out as a leader!
Copilot for Business 🧑💼
GitHub Copilot, the AI-powered developer tool that suggests code in real-time, is now available for every developer, team, organization, and enterprise with its new offering, Copilot for Business. This tool has transformed developer productivity for over one million people and can help code up to 55% faster. With new capabilities such as a more advanced OpenAI model, AI-based security vulnerability filtering, and VPN proxy support, this tool can offer even greater benefits to organizations.
Tools of the Trade ⚒️
ChatGPT for Coding 👩💻
Replit has launched Ghostwriter, an AI chat bot to help programmers write and debug code. The tool will tell programmers when they have made a mistake and suggest ways to fix it. The chat bot, which the team hopes will enable skilled coders to create billion-dollar companies by increasing their ability to build complex software programs, is integrated into Replit's IDE.
![[video-to-gif output image]](https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7bb5e38c-7506-4164-bab5-2868e5deb81c_800x450.gif "[video-to-gif output image]")
Git Your Commands in Order 🧑⚖️
Git Command Generator is an AI-driven web app that helps users quickly find the appropriate Git command they need by entering a description of what they want to accomplish. Based on their input, the app suggests relevant Git commands that users can then copy and paste into their terminal or command line interface, saving them time.
Research - Easy/Peezy 🤓
ResearchGPT allows users to converse with a research paper. The app extracts text from a PDF and creates embeddings to generate a response to a user's question using the OpenAI API. The app also provides a source for the text used to generate the response and the page number.
![[speed output image]](https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fee7846e4-f0da-4c98-820e-217d736fc3fc_446x313.gif "[speed output image]")
The AI Marketeer 🛒
Boost your local sales with unbelievably easy video using AI with Waymark. Waymark lets you get from ideas to a top quality ad in just three steps - Import business information, AI produces your video, and polish your video with any last adjustments, then share, download, and get ready to air!
Hot Takes 🔥
Beginning of AGI 🦾
And In case AGI takes over 🤖
Cost of AGI 🤑
AI Meme of the Week 🤡
That’s all for this week!
Will see you next Saturday with more such content. Don’t forget to subscribe and give your feedback below.
BONUS 🎉
Share this newsletter with three other friends and stand a chance to win my book GPT-3: The Ultimate Guide to build NLP Products with OpenAI API. 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!