I spent so much time waiting for delayed flights since the official release of Meta’s new Llama 3.1 that I had plenty of opportunities to run some tests. However, the quality of the model is not the key insight; instead, what you want to think about is Zuck’s open source playbook.
In practical terms, Llama 3.1 405B is an open source foundational model that rivals OpenAI’s closed source GPT-4. Don’t worry about why this is true or any of the technical details. They don’t matter for this discussion. What matters is that 405B is about as good as the best flavor of GPT-4.
Importantly, Llama 3.1 comes in two smaller sizes: 70B and 8B. You can run 70B on a desktop workstation or high-end gaming PC, and 8B can run on high-end smartphones. If you believe that the future of AI includes small models running at the edge (nearer to users), the advent of very high quality, open source models that can easily be post-trained by developers is a requirement.
As for Zuck’s open source playbook? The strategy is savage: make everything free to use, make it an awesome developer experience, hook everyone, incentivize partnerships, organically grow an ecosystem, own the category. He did it with React. He did it with PyTorch. He’s doing it with Llama.
The key takeaway? Going forward, it’s going to be very hard to charge for access to closed source foundational models. Six months from now, we are very likely to look back at the release of Llama 3.1 as a defining moment in the business of AI. Zuck never ceases to impress.
Author’s note: This is not a sponsored post. I am the author of this article and it expresses my own opinions. I am not, nor is my company, receiving compensation for it. This work was created with the assistance of various generative AI models.