AI-dapting to the New Age

One stop-shop to get up to speed on all the latest innovations in AI

Hey there folks 👋

Welcome to the inaugural issue from Unwind AI.

If you want to know what all has been cooking in the AI and NLP field, you’re just at the right place my friend! Every week you’ll be served with what’s new and trending in the area, be it tools, tricks, articles, research, opinions, and basically everything!

This issue covers:

  1. Latest Developments 🌍

  2. News from the Industry 🧑‍🏫

  3. Hot Takes 🔥

  4. AI Meme of the Week 🤡

So without further ado, let’s get into it.

Latest Developments 🌍

New 🌟 to the OpenAI ecosystem

This week, NLP giant OpenAI introduced a new model “text-davinci-003” to the existing model family in GPT-3. Built on the same InstructGPT, it is supposed to be better at:

  • Producing clearer and more engaging content

  • Handling complex instructions

  • Longer content generation

Here are some examples 👇

🕹️ Try it out yourself - OpenAI API 

OpenAI ChatGPT: Language Models for Dialogue 🤖

OpenAI released a sibling model to InstructGPT, which is trained to have a conversation like humans by being able to answer follow up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.

Let’s quickly look at its capability in a snapshot:

Given a question, ChatGPT provides you with three things:

  • A simple answer

  • Evidence supporting that answer (code snippet in this case)

  • Follow up explanation to the given rationale for further inquiry

🕹️ Try it out yourself - ChatGPT Demo

CICERO - AI that can Negotiate

CICERO by Meta AI is the first AI agent to achieve human-level performance in the complex natural language strategy game like Diplomacy which wasn't even conceptually possible for machines to do. It combines two different areas of AI: strategic reasoning and natural language processing. It has the three key components required for any diplomatic conversation:

  • Cooperate

  • Negotiate

  • Coordinate

Prompt Chaining with LangChain AI

LangChain AI allows logical connection of prompts in a way that prompts can talk to each other that too out-of-the-box. Now you can easily chain multiple prompts together to build sophisticated LLM applications.

LangChain is a python library designed to provide:

  1. Comprehensive collection of pieces you would ever want to combine

  2. Flexible interface for combining pieces into a single comprehensive “chain”

  3. Schema for easily saving and sharing those chains.

What can you do with LangChain AI?

  1. Solving complex word Math Problems

Image
  1. Reasoning between Prompts (Q/A Example)

Image
  1. Chain Reaction (Step-by-step derivation)

Image

If you liked this reason-based prompt approach, support the project with a 🌟.

Opensource GPT-JT FTW

GPT-JT by Togetheris is a fork of GPT-6B, fine-tuned on 3.53 billion tokens, that outperforms most of the 100B+ parameter models at classification. It is claimed to be an opensource AI equivalent to OpenAI’s “text-davinci-002”.

It is built on GPT-J-6B created by EleutherAI and highlights the key strength of opensource AI, where community projects can be incrementally improved and contributed back into opensource resulting in a value chain that fuels innovation!

🕹️ Try it out yourself - GPT-JT Demo

AI-powered Image Editing

Introducing Text-Based Inpainting with CLIPSeg x Stable Diffusion on Hugging Face spaces by Niels Rogge.

All you need to do is:

  • Upload an image

  • Type what you want to replace

  • Type what you want to replace it with

Voila, sit back while AI will edit the image! 🛀

Image

News from the Industry 🧑‍🏫

Generative AI Landscape 🚀

Here are the companies leading the Generative AI revolution from the front:

AI: The Beast or Jerusalem?

One of the most interesting conversations I've watched recently on the current state of AI. Watch Tenstorrent’s Jim Keller with Jonathan Pageau dive into the world of AI discussing the pros & cons of technical advancements.

Hot takes 🔥

Will GPT-4 be the holy grail of AI? 🤯

If it launches with ~100 Trillion parameters, it will be able to spin off multiple GPT-3 like models out-of-the-box.

Image

Remembering Matlab 🔬

GPT-3 and Drama 🎸

AI meme of the week 🤡

Image

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 - your small contribution is what keeps me going!

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!

Image

🎁 Every paid subscriber will also receive $39 USD worth of learning resources on trending topics like Python, Data Science, Machine Learning, and NLP!

Reply

or to participate.