🤝 AidfulAI Newsletter #13: Meta Released With Llama 2 the Future of Privacy-Friendly Chatbots
Dear curious minds,
Welcome to this week's AI newsletter. We've got updates from big players like Elon Musk's xAI and Google's Bard, innovative tools like Stability AI's Stable Doodle and Meta's Llama 2, and the exciting AI-first notebook, NotebookLM from Google. Plus, a noteworthy podcast about the ChatGPT Code Interpreter. Dive in for your dose of the latest in AI!
Major AI News
🤖🌌 Elon Musk Unveils xAI: Goal To Build A Good AGI And Understand The Universe
Elon Musk presented the team and mission of xAI in a Twitter space.
You can listen to the recording shared by @xai.
A nice summary of the event is stated here.
My take: It is good to have more competition in the LLM landscape, and the founding members have some really impressive backgrounds. Looking forward to their first release, which is already announced in a couple of (Elon-)weeks.
🌍🎁 Google's Chatbot Bard: Regional Expansion and New Features
Bard is now available in over 40 languages, including Arabic, Chinese, German, Hindi, and Spanish and rolled-out to many more countries.
Google described in their blog article features they added to their ChatGPT competitor Bard.
Listen to Bard's responses, and change the tone and style of Bard's responses.
Pin and rename conversations with Bard, export Python code to Replit in addition to Google Colab, and share responses with friends.
Use images in your prompts.
My take: Nice to have these new features for Bard. Adding an image to your prompt was announced first by OpenAI, but Microsoft and Google delivered first.
Privacy-Friendly AI
✏🖼 Stable Doodle: Stability AI's Tool Turning Sketches Into Dynamic Images
Stability AI launches Stable Doodle, a sketch-to-image tool that turns simple drawings into dynamic images.
Stable Doodle is intended for both professionals and beginners, allowing anyone with basic drawing skills to generate high-quality original images quickly.
An ostrich in Origami style created with Stable Doodle on Clipdrop. Up to 14 different Stable Diffusion XL styles can be chosen for the generated images, ranging from realistic to creative.
Stable Doodle is available for free testing on the Clipdrop by Stability AI website.
My take: Not the first approach which converts simple Doodles into images, but definitely a nice and easy to use one. If you have a child, try it together and have some fun exploring it.
🦙🦙 ChatGPT competitor: Llama 2 released by Meta
Llama 2 comes with a license for research and commercial use (in contrast to the research-only license for Llama v1).
Partnership with Microsoft, which includes Llama 2 in its Azure AI catalog.
7, 13 and 70B base (completes text) and chat (fine-tuned for conversations and instruction following) versions released.
34B versions trained, but not released, but not released yet due to a worse rating in proactive risk identification.
The 34B model shows a higher violation percentage and is not released yet. 70B chat version now in 2nd place on 🤗 Open LLM Leaderboard, as a fine-tuned Llama 1 based model sneaked itself on the first place.
Current snapshot of the Open LLM Leaderboard Try it on HuggingChat: Llama-2-70b-chat-hf
My take: Awesome to finally have an open-source model on GPT-3.5 level. Releasing it for commercial use will skyrocket improvements and fine-tunings based on Llama 2 even more than for the initial version.
PKM and AI
📑🤖 NotebookLM: Google's AI-first notebook app
Google announced at their Google I/O conference an AI-first notebook as Project Tailwind.
The project is now rolled out as NotebookLM, starting with a small group of users in the USA.
NotebookLM has been designed to help users deal with the rapid growth of information and the challenge of synthesizing facts and ideas from multiple sources.
It acts as a virtual research assistant that can summarize facts, explain complex ideas, and brainstorm new connections based on the user's selected sources.
Unlike traditional AI chatbots, NotebookLM allows users to “ground” the language model in their notes and sources, effectively creating a personalized AI that's versed in the information relevant to the user.
It generates for uploaded Google Docs a summary and suggests key topics and questions for a better understanding.
After uploading the document shown in the center, the guide on the right side is generated by AI. source After uploading the document shown in the center, the guide on the right side is generated by AI. source
My take: An AI-first note-taking app sounds interesting, but I would rather not share all my thoughts with Google.
Podcasts
⌨🚀 ChatGPT Code Interpreter gets more awesome if you know how to use it effectively!
Outstanding podcast episode about the Code Interpreter by the latent.space podcast.
Code interpreter is not a coding, but an LLM revolution. Listen to understand why!
On top @swyx created the living manual of the Code Interpreter.
My take: Uploading files and a self-correcting loop for working with these is a game changer.
Disclaimer: This newsletter is written with the help of AI. I use AI as an assistant to generate and optimize the text. However, the amount of AI used varies depending on the topic and the content. I always curate and edit the text myself to ensure quality and accuracy. The opinions and views expressed in this newsletter are my own and do not necessarily reflect those of the sources or the AI models.