Llama, llama, llama 🦙

In case you missed it, last week Meta created a splash by releasing Llama 3 — a series of best-in-class open weight large language models (LLMs) that are going to have a huge impact on the AI landscape.

Meta has really raised the bar, delivering open weight models that are on par with the best commercial offerings available today. The 70B parameter model has demonstrated results that are close to GPT-4 on a wide range of tasks, while the smaller 8B model outperforms GPT-3.5. What's even more exciting is that Meta has 400B+ models still in training, which might even surpass the current top models from industry leaders like OpenAI and Anthropic.

The long-awaited release of Llama 3 has significant implications for enterprises. It levels the AI playing field and opens up new strategic options that were previously unavailable or cost-prohibitive. Rather than being solely dependent on commercial API providers, companies now have the option to self-host powerful open source models that compete with the leading commercial models. This provides greater control over deployments, model fine-tuning, data privacy, and costs at scale.

The decision by Meta and Mark Zuckerberg to open source Llama 3 raises interesting questions about their underlying motives. On the surface, it may appear to be a purely altruistic move, driven by a desire to democratise access to cutting-edge AI technologies and foster innovation in the wider community but there are obviously strategic considerations at play as well.

By open sourcing Llama 3, Meta is positioning itself as a leader and key enabler in the AI ecosystem. This helps to attract top talent, build goodwill among developers and researchers, and shape the direction of future AI development. It also allows Meta to benefit from the collective intelligence and contributions of the open source community, accelerating the pace of innovation and improvement of their models.

Meta has also effectively pulled the rug out from under companies like OpenAI and Anthropic. If everyone has access to the same foundational technology, it becomes much harder for these companies to differentiate themselves, maintain a foothold in the market, and ultimately disrupt Meta.

While this interpretation may seem cynical, it's not without precedent in the tech industry. We've seen similar tactics employed in other domains, such as open sourcing key infrastructure components to undermine competitors and establish de facto standards.

There may be elements of altruism in Meta's decision to open source Llama 3 or it might be part of a broader strategic play. Either way, it doesn't really matter. Llama 3 is out there now, and it is going to significantly change the way companies deploy AI use cases.

Most enterprises currently rely on commercial APIs for their AI use cases, where the benefits of managed services outweigh the costs and limitations. With the availability of powerful open source models like Llama 3, many AI professionals I've spoken to are now seriously exploring self-hosting, especially for more experimental and sensitive applications where data privacy, customisation, and cost control are paramount.

While self-hosting Llama 3 models offers benefits, it is not without its challenges. It requires significant compute infrastructure, ML engineering talent, and robust governance and safety controls. Companies need to carefully weigh the upfront investment and ongoing overhead against the long-term flexibility and potential cost savings. For many organisations, the managed convenience and simplicity of commercial APIs may still be the best choice.

Whether you adopt a self-hosting strategy or not, the release of Meta's Llama 3 is a game changer for the enterprise AI landscape. It demonstrates that you shouldn't underestimate the rapid progress of the open source AI community and it provides enterprises with powerful new options for building and deploying AI solutions. The future is bright and I for one am excited to see where Llama 3 and its successors lead us.

Game on.

Want to chat to Euan about AI? Click the button below to arrange a call.

Euan Wielewski is an AI & machine learning leader with deep expertise of deploying AI solutions in enterprise environments. Euan has a PhD from the University of Oxford and leads the Applied AI team at NatWest Group.