Meta just launched Llama 3, claiming it as the most powerful open-source language model out there. It’s a huge step forward for AI to understand and communicate almost like humans. Llama 3 was trained on some text and code data, so it’s great at handling complex instructions and conversations.

This is going to open up many exciting possibilities for all kinds of applications, so we decided to give it a SWIRL.

I downloaded it after reading about Llama3 and how well it compares to GPT-4. And then I hooked it up to SWIRL. Here’s what I did:

First, I edited the AI Provider for Llama2, changing it to Llama3. SWIRL, echoing the tenets of AI Infrastructure Software, allows for instant LLM or AI Model swapping.

And, then RAG the results in an instant.

SWIRL follows this model: configure, not code. It comes with out-of-the-box support for a huge list of known Generative AIs and embedding providers. I made it active and set it as the default for RAG.

But to compare Llama with GPT-4, I have to do my favorite searches: “Why is Elon Musk unpopular?” And “Tell me about SWIRL AI Connect.”

Here are the results:

Llama3 is impressively close and much more coherent than Llama2!
The race continues… and maybe that’s the point.

There will always be another amazing Generative AI. But your data can stay where it is—until it’s time to RAG it. Then you can just use SWIRL.