Meta's Llama-v2: A New Dawn for AI?

Welcome to another edition of the best damn newsletter in AI. Here we’ll break down AI topics that matter, open your mind to use cases, and keep you ahead of the curve.

Our #1 goal is to be useful. So please shoot us an email 📩 if you have questions or feedback, and especially if you implement something we share!

Here's what we're covering today:

• Llama-v2 explained

• Microsoft, Apple and Google are all making moves with AI-powered products

• How to make sure you stay compliant as you build AI products and internal solutions

Let’s get to it! 👇

TODAY'S PERSPECTIVE

Why the new Llama-v2 model changed the game (explained!)

There are a lot of moments that define history that most people don't notice. And at the risk of sounding cheesy - Tuesday might have been one of them.

On Tuesday, Meta released Llama-v2, the next iteration of their open source large language model to compete with ChatGPT.

But in contrast to their release of Llama 1, where it became fully open source seemingly on accident (someone leaked it...)

This time, Meta said go for it. They made the full AI model available and free to everyone. AND, gave us the commercial rights to it. For anyone.

You read that right.

That means anyone can now download an AI model trained by one of the largest tech giants and incredible AI teams. And then use it in their products, services, internal processes and more.

This sets crazy precedent for what Meta might do next. After all - they are in an interesting position where they have one of the leading AI research teams in the world; yet no business model around large language models. So they can afford to “scorch the earth” so to speak and turn LLMs into a commodity offering.

Gimme the details

• Llama 2 was trained on a dataset of ~1.5B words (40% more than their last model).

• It comes in 3 sizes, which is helpful for speed - the smaller ones are faster but the bigger ones are more powerful.

• They have a base model version (no fine tuning) and chat versions that have been fine tuned to be good at chat conversations (like ChatGPT)

How to get access

Downloading and using a large language model requires a lot of compute (and fancy hardware) because well… it’s large.

We recommend testing out Llama for yourself at one of the currently free websites hosting it for you:

• llama.perplexity.ai is a simple chatbot powered by the smallest of the Llama models

• llama2.ai was created by a16z and Replicate and gives you all the settings and different model sizes

• Hugging Face’s Hugging Chat

If you are a business, you might also consider setting up an account and starting to play with your own instance of it on:

• Databricks

• Banana.dev

• AWS

• Azure

• Baseten

If you still want to, you can request a link to download the AI model itself at https://ai.meta.com/llama/ 

FROM OUR PARTNERS

Use AI to Prevent Compliance Headaches

Even if it's nerdy - compliance and data privacy are necessary and crucial in AI.

That's why we've been crushing on Secureframe - The automated tech compliance platform built by compliance experts.

Secureframe helps you get and maintain compliance requirements like SOC 2, ISO 27001, GDPR, and more. And they use AI to streamline those processes, making it even easier for you.

If you're building an AI product or integrating AI into your company, check out how Secureframe can make your compliance a breeze.

And for premium members, we have a hands-on workshop with Secureframe today Thursday, July 20 from 1-2pm PT. And it'll be a perfect way to learn about what compliance and data privacy considerations you should have as you're building or using AI. Grab your spot.

GET ON THE LIST FOR UPCOMING COHORTS 🔥 

You might have missed it because it happened so fast 😅 

But we launched our first cohorts of our 3 Week Prompting for AI Operations Bootcamp a few weeks ago (and then quickly sold out in a few hours…)

We have SO appreciated all the messages about it and are working on scheduling more cohorts! If you’re not already on the waitlist, grab a spot here.

Premium Members: don’t worry, you’ll get first access once we open spots again! (p.s. a good reason to join if you’re not already a subscriber!)

LINKS

For your reading list 📚

Big tech keeps experimenting with AI…

• Reports say that Apple is experimenting with their own AI model, nicknamed “Apple GPT”, but that they don’t know what to do with it

• Google is testing AI-powered video backgrounds for Google Meets

Microsoft furthers its moves for the enterprise…

• If you want a version of ChatGPT that doesn’t share your data back with the model provider - Microsoft just announced Bing Enterprise. Secure for only $5/user/mo

• Microsoft also announced pricing for Copilot, and people are calling it steep at $30/user/mo

Important stories about job loss and AI that are unfolding…

• This story highlights a call center employee’s battle with AI as more of the customer support role becomes automated

• And we already covered it, but the Hollywood strikes around AI are also important to be paying attention to

And if you’re really nerdy…  

• A new research paper proved what people have been complaining about (at least a little bit)… Prompt Drift is real and both ChatGPT and GPT-4 have gotten worse on some aspects since March

Premium Members: if you want to read more about Prompt Drift and how to handle it, read our chapter on it here.

That's all!

We'll see you again on Tuesday. Thoughts, feedback and questions are much appreciated - respond here or shoot us a note at [email protected].

Cheers,

🪄 The AI Exchange Team