😸 What Meta’s move means

PLUS: You can fine-tune GPT-4o Mini for free!

Welcome, humans.

If you’re an AI or software company, we have 450K AI enthusiasts ready to test your product. A few sponsorship slots opened up in August, and now we’re offering them on a first-come, first-served basis. Fill out our sponsorship form here to claim your spot before they fill up.

And meow, before we begin, may I present to you the cutest thing on the internet today:

Here’s what you need to know about AI today:

Meta’s Llama 3.1 is out, and it’s the first open-source model on par with ChatGPT-4 / Claude 3.5 Sonnet.
U.S. Senators want OpenAI to share its new foundation models with government officials.
OpenAI made fine-tuning ChatGPT-4o Mini free until September.
A Danish survey of 100K workers revealed widespread ChatGPT adoption.

Meta’s Llama 3.1 is here, and it’s competitive with the best AI models.

The Neuron is not generated by AI and will be accurate and appropriate, unless there’s a good joke to tell…

It’s official: Meta has released Llama 3.1, a family of 3 new open-source AI models that anyone can download and use to build new AI apps.

The best version of the model, the 405B version, is available right now on WhatsApp and the Meta AI chatbot, with Instagram and Facebook coming soon.

Meta's biggest model is a big deal. It's the first open-source model to rank among the top tier of AI models. Yes, that’s the ChatGPT-4s, the Claude 3.5 Sonnets, and the Gemini Ultras, which are closed models, meaning no one can see how they’re made.

Previously, Meta's models were good for being small + open source, but not "frontier" (best in class) models. This time, the 405B performs comparably to GPT-4 and Claude 3.5 Sonnet.

It's strong in math, common sense, and has a context window of 128K tokens (around 200 pages of text).
Even the smallest Llama 3.1 model, 8B, outscores OpenAI’s-4o mini!

Because open-source means anyone can see how the model works, it means anyone can examine, modify, and build on top of the code.

It’s like having the recipe to the best dish in the world. You don’t have to pay a restaurant to make you the dish—you can make it yourself, modify it for dietary restrictions, or add a few extra flavors.

Llama 3.1 is the world-class dish.
Its downloadable code is the recipe.
All the new AI apps built on top of it (and there will be a lot) are the modified and optimized dishes.

Also, Meta’s AI can now generate custom images. Just type “transform my cat into an NBA star” on Facebook, and it'll generate the images in real-time while you prompt it. More cat photos on Facebook? Sign. Us. Up.

Mark Zuckerberg wants Meta AI to be the most popular AI chatbot by year-end, and he’s planning to do so by force-feeding Llama 3.1 to his billions of users across Facebook, IG, WhatsApp, Messenger, and, yup, AI glasses.

Our take: Meta's already slapping its AI onto Instagram and Facebook as a way to search for things, but most of our friends find the AI search bars really annoying. Hopefully, they'll fix the UI, or soon, we'll have AI that’s so advanced it can do everything except help you find your friend's profile.

Maybe they should call it the “Hide and Seek” bar instead!

Which Meta AI feature has you most intrigued?

FROM OUR PARTNERS

Want to tune your own models to outperform ChatGPT?

If you want to build your own AI models that surpass ChatGPT's performance, you need OctoAI's strategies.

Join OctoAI's live session on August 7th, and you’ll learn how to optimize Llama3-8B for a 47% quality boost while cutting AI costs by 90%.

You’ll also learn:

Optimal model performance techniques
Efficient Llama 3.1 fine-tuning strategies
Continuous development for LLM apps

They’ll cap things off with a live demo comparing Llama3-8B to GPT-4, followed by an expert Q&A.

When you register, you receive the slides, code samples, and a session recording—even if you can't attend live.

Sign up now.

Around the Horn.

U.S. Senators want OpenAI to share its new foundation models with government officials before publicly releasing them
A Danish survey of 100K workers revealed widespread ChatGPT adoption, with many estimating AI tools could halve work time for ⅓ of tasks.
Legal startup Harvey raised $100M at a $1.5B valuation to expand its copilot for lawyers.
In addition to using it on Meta, you can also try Llama 3.1-8B and 3.170B on Groq, Amazon Bedrock, and Perplexity right now.

Treats To Try.

*Contractor+ is helping solve the 4.5 million US housing deficit with AI-powered estimates and localized pricing. Invest as little as $100 here.
Adobe released new FireFly tools for Illustrator and Photoshop, allowing designers to generate detailed textures, shapes, and images w/ text prompts directly in the app.
Level AI automates quality assurance and enhances agent performance for customer service centers (raised $39.4M)
HackerPulse is a dev profile aggregator that pools together all your data from multiple platforms (GitHub, StackOverflow, LinkedIn, HackerNews).
Hotseat AI quickly finds relevant quotes in complex legal documents using semantic search.
ChartFast automates data analysis and visualization, creating complex graphs w/o manual work.
Quick Headshots transforms selfies into a batch of 80-400 professional AI-generated headshots within 2-5 hours.

*This is sponsored content. Advertise in The Neuron here.