Keep an eye on AI
What's New and Next in the AI space!
Hey there folks 👋
Welcome back to the third weekly issue from Unwind AI! It's been an exciting few days and I've been delighted to see so many new subscribers join this publication. I hope you find the content to be informative, thought-provoking, and inspiring.
I have some exciting news and updates as well as interesting tools and perspectives to share with you all. This issue covers:
Latest Developments 🌍
News from the Industry 🧑🏫
Tools of the Trade ⚒️
Hot Takes 🔥
AI Meme of the Week 🤡
What are we waiting for? Let’s dive straight into it!
Latest Developments 🌍
Robotic Transformers 1 by Google 🤖
Google has developed a new opensource AI model, called the Robotics Transformer-1 (RT-1), that can perform a variety of tasks, including grasping objects and playing catch. The model is based on the transformer architecture and can take in a sequence of images and a natural language instruction to output robot commands. RT-1 is trained on a large dataset of tasks and is said to be better at adapting to new tasks, environments, and objects than previous techniques.
Dramatron by DeepMind 💃
DeepMind has launched Dramatron, an AI script-writing tool that leverages large language models to generate immersive scripts. It is a co-writing tool that can help authors to write theatre scripts and screenplays. The tool uses hierarchical story generation to create consistency in the output, which is then given to human writers for editing and rewriting. The idea is to use the tool for "world-building" and idea generation rather than writing full plays.
Multilingual AI Models by Cohere 🈂️
Cohere introduces its first multilingual text understanding model that supports over 100 languages and delivers three times better performance than existing models. The model is specifically trained for search, content aggregation and recommendation, and zero-shot cross-lingual text classification. This enables a range of use cases for multilingual settings, such as improving search results and identifying harmful content in online global communities.
Some of the many usecases of this model are:
Multilingual Semantic Search: Keyword search has its limits, which can often lead to irrelevant results. Semantic search powered by Cohere's multilingual model can improve search quality & works across languages.
Customer Feedback Aggregation: Extracting insights from product feedback in multiple languages is now possible with Cohere’s multilingual model that can map text in different languages to the same vector spaces, allowing users to derive insights across languages.
Zero-Shot Content Moderation: Tackle the challenge of content moderation across 100+ languages with Cohere's multilingual embedding model! Only requires training data in one language to classify harmful content in 100+ languages.
All you need to try this interesting tool is cohere's API key and a few lines of Python code. Get the API key here!
Train your own InstructGPT 😲
CarperAI has developed Transfer Reinforcement Learning X (trlX), an opensource library that enables the training of language models with Reinforcement Learning from Human Feedback (RLHF). RLHF is a method used to create safe and responsible machine learning (ML) models by aligning them with human preferences. This technique has been used to train language models that are better at following user intentions than GPT-3, while also making them more truthful and less toxic.
trlX allows for fine-tuning of HuggingFace-supported language models using reinforcement learning via either a provided reward function or a reward-labeled dataset. It can be used to train large language models with up to 20B parameters and supports distributed and scalable hyperparameter optimization using Ray Tune and Weights & Biases.
Stable Diffusion for Music 🎼
Riffusion is a music-generating tool that allows you to explore a wide range of musical styles, instruments, genres, and sounds. It has been built by fine-tuning the Stable Diffusion model to generate images of spectrogram which are then converted to audio clips.
You can use text prompts to generate music, while you can also play with the settings to explore the latent space of sound. The tool also has an img2img diffusion feature, which keeps your riff on the beat and impacts melodic patterns. By adjusting the denoising setting, you can create more creative output and get offbeat.
News from the Industry 🧑🏫
AI Output Detector
Now AI can detect whether the output is generated by AI. The output detector model uses 🤗 transformer’s implementation of RoBERTa. All you need to do is to enter some text in the textbox and get the predicted probabilities of text generated by AI or humans. You can play with the demo on HuggingFace Hub.
Software Development Revamped 🏗️
A website was built in less than 10 minutes using the ChatGPT language model and Replit. ChatGPT provided the initial HTML and CSS for the landing page, which was then run on Replit. Additional elements and visual upgrades were added by prompting ChatGPT for the necessary code. The resulting website, although not the final version, was a functioning "startup" that was live in seconds.
Tools of the Trade ⚒️
GPT-based Browser Assistant 🌐
Based on GPT-3 model, XP1 is a productivity assistant designed to help users with tasks such as summarizing documents, extracting data from text, and generating personalized emails. It uses the text from the user's open browser tabs as context for its responses. It can be accessed via a keyboard.
Generate Photorealistic Image 📷
Lexica Aperture is a model that can generate realistic-looking photographs. It is trained on high-resolution photos, so the results it produces are photorealistic. The model allows users to generate photographs based on text prompts, making it a powerful 3D engine. The beta version is available for users to try out; simply log in and click the Aperture tab.
Check out this image generated by Lexica Aperture using the prompt “Polaroid portrait of JFK posing in the White House Oval Office with Marilyn Monroe, color polaroid, historic photo”.
Ellie - your Email Assistant 💌
Powered by OpenAI, Ellie is an AI email assistant that is currently in beta. It learns from your writing style and generates replies as if they were written by you. It does not read your emails, but you can provide it with some example email data to better match your style. There is a limit on the number of replies, but you can sign up for the Early Bird plan to get more!
Unique Logos by AI 🎨
MakeLogoAI is an awesome way to get a unique logo for your startup with minimal effort. Simply tell input your startup name and what it does and AI will generate multiple logo concepts for your business, for a fraction of the cost of hiring a professional designer. The logos will be delivered in HD and with a transparent background, so you can use them in any way you want.
AI Writing Assistant for Notion 👩💻
Webcopilot helps you write your Notion pages more quickly and efficiently. It uses advanced AI technology to help you save time while producing quality content. The AI Writing Assistant helps you structure your Notion pages and provides you with helpful prompts and suggestions. All of this is available for free, forever!
AI Interior Designer 🛏️
Interior AI is a platform that allows users to create new looks and features for their interior spaces using a 2D image of the space as input. The platform offers 32 pre-selected designs like Christmas, minimalistic, bohemian, etc., and allows users to change the function of different rooms, such as the kitchen or living room. You will get 5 free renders at the beginning. To use Interior AI:
upload your room’s image
choose a room, mode, style, number of renders, quality, and privacy setting, and
click "render"
And Voila! Your house is revamped.
Hot Takes 🔥
ChatGPT or an Assignment Buddy 👩🎓
Lawyer or No Lawyer 👨⚖️
ChatGPT: Factual and accurate? Are you sure? 🤔
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!