- Stable Confusion
- Posts
- AI Report #10: It's Llama Time
Hey folks! It’s been a while.
There have been a ton of announcements in the last week alone, so let’s get you up to speed.
Reading time: 5 minutes
📰 News
Llama 2 is here
Meta and Microsoft co-announced the release of Llama 2, the newest iteration of Meta’s open source LLM. Here are the highlights:
It’s open source and free for research and commercial use
The model was trained on 2 trillion (!) tokens
It’s available in three sizes: 7 billion, 13 billion, and 70 billion parameter versions
It’s committed to being a responsible use of AI, and has the backing of major tech companies like Microsoft, Nvidia, AWS, IBM, and others
You can read get all the details (and even download it yourself!) from here or on HuggingFace. You can also read the full research paper here.
Elon announces new AI company, xAI
Announcing formation of @xai to understand reality
— Elon Musk (@elonmusk)
4:23 PM • Jul 12, 2023
Elon Musk announced the launch of his latest venture, xAI. The goal? To “understand the true nature of the universe”.
The company is separate from Twitter (and its parent company, X Corp), and already has impressive team of researchers from OpenAI, Google, Tesla, and the University of Toronto.
They’re also accepting applications here if you’re interested (not a sponsored post btw).
Anthropic announces Claude 2
OpenAI competitor Anthropic just announced the release of Claude 2, their version of GPT-4. They highlight that it scores high on different exams, while also being “easy to converse with”.
Anthropic also released a ChatGPT-like web interface where you can talk to Claude, which is available now to users in the US and UK. The research paper, with full technical details behind Claude, can be found here.
Other quick updates
OpenAI releases Code Interpreter feature for paying users. It’s nuts; it can allow ChatGPT to run code and allows users to attach files. It’s performance is so good that some are even saying it may have GPT-4.5 running on its backend.
Meta introduces CM3leon, a state-of-the-art generative art model
FTC begins investigation in OpenAI
Tools and Resources
How to build a summarization extension for Chrome using Cohere
GPT Prompt Engineer - prompt like a pro with this open source code
Alpha - an AI-powered investing co-pilot (sign up for Alpha access)
Coolest thing I’ve seen all week
Stable Doodle. Stability AI, the folks behind Stable Diffusion, just released a doodle-to-image model that is accessible through their webpage. Draw a pic, give a prompt, and it will output 3 generated images. Super neat.
Here’s my attempt at a comicbook-style Spongebob:
Top left is my drawing, the other three were generated with Stable Doodle.
Pretty impressive. The ones by the AI are pretty good, too. 😎
That’s all for today, thanks for reading!
What did you think of today’s newsletter? Reply to this email to let me know what you think or what you want to see more of.
Cheers,
P.S. Sharing is caring. Consider sharing this newsletter with a friend.