- The Renaissance Times
- Posts
- Deep Dive: Meta's Llama 3
Deep Dive: Meta's Llama 3
The Renaissance Times
April 19, 2024
Dear Artisan,
We’ve had a day to digest Meta’s latest AI announcements, listen to a new Zuck podcast (and his fake AI facial hair), and we are live with our first look on Llama 3.
Let’s dive in.
Hot Off The Press
OpenAI Makes First India Hire in Bid to Shape Regulation Early
Meta AI is restricting election-related responses in India
KissanAI Releases Dhenu Llama 3, an Indic LLM for Farmers
Netflix’s What Jennifer Did used AI-generated images in scenes describing the subject
Hugging Face releases a benchmark for testing generative AI on health tasks
The One Big Thing
Looks like Meta is serious about this AI thing.
Llama 3’s initial results look quite impressive. While we know that the model is not as large as competitors from Anthropic and OpenAI, it has been able to overcome this with large amounts of data and aggressive fine-tuning of the model.
I think Meta and Llama-3 is the final nail in the coffin to several misconceptions I've been fighting against for the last year.
Llama-3 Chat was trained on over 10M Instruction/Chat samples, and is one of the only finetunes that shows significant improvements to MMLU.… twitter.com/i/web/status/1…
— Teknium (e/λ) (@Teknium1)
3:35 PM • Apr 19, 2024
10 million human-annotated examples are expensive. This is thousands of humans spending hundreds of hours fine-tuning datasets to prepare them and optimize the model as well as possible.
These are the things only a company like Meta can do due to the sheer capital requirements. Which has led many people to ask - why would you spend billions of dollars to create this sort of powerful model and open source it?
Aren’t you killing your entire moat? Maybe not.
i’ve been thinking about Meta’s support for open-source AI all day, wondering what zuck’s business justification must be. you can’t run a $1T co on ideology alone.
but when you google the financials of Meta ($134B/yr) compared to a co like OAI, ($1.6B) it starts to make sense:
— Carmen Gutierrez (@carmguti)
7:48 PM • Apr 18, 2024
Now, to the results themselves.
As we showed yesterday, Llama 3 is already the best-performing open-source model on the market right now. Now, it looks like it’s on par with GPT-4 Turbo.
Uhh excuse me wtf LLAMA3 ranking 1st????? in lmsys arena in English? Kudos to team @AIatMeta , based AF 👏👏 for open sourcing literal GPT-4 level model, (almost) no strings attached🥳
— Simo Ryu (@cloneofsimo)
11:45 AM • Apr 19, 2024
The coolest part of all this is running the model on the fastest custom hardware out there, Groq.
This thing is fast.
My mind is blown.
@GroqInc is serving LLaMA 3 at over 800 tokens per second!
800. Tokens. Per. Second.
This unlocks so many incredible use-cases.
It's one thing to see my demo — it's another thing entirely to experience it for yourself.
Do yourself a favor and try it asap. twitter.com/i/web/status/1…
— Matt Shumer (@mattshumer_)
4:13 PM • Apr 19, 2024
The Gallery
Auto-generate web forms from a document
Posting this again as it got a fair bit of interest last time 😜
It’s an AI + Forms experiment I made 🤖
You give it an image of a document form and it generates a web version
— Tim Paul (@timpaul)
4:54 AM • Apr 19, 2024
Tools
Must have tools for every artisan to add to their toolkit:
Direqt: A chatbot platform for publishers reader engagement through personalized conversations on websites.
Jamie: A tool to automate meeting note-taking by providing summaries, transcripts, and action items from audio recordings.
Docus: A chat assistant for personalized health consultations and second opinions.
UI Bakery: A tool to create web applications from text prompts using low-code development.
DocsHound: A tool to automate documentation creation and maintenance.
Closing Thought
Zuck with beard is a scary speciment
Work With Us!
The AI Renaissance is coming and we are building the best community of the people making it happen.
Contact us to sponsor your product or brand and reach the exact audience for your needs across our newsletter and podcast network.