Evaluating Meta’s New Llama 3 Models

By Adam Pease

Evaluating Meta’s New Llama 3 Models

Meta’s release of Llama 3 has been met with positive acclaim for its competitive capabilities. This blog discusses the new series of models and Meta’s overall approach to AI.

Meta’s Llama 3

Meta has released the latest in its series of quasi-open-source language models, Llama 3. The model is designed primarily for chat use cases and comes in a smaller 8B parameter and larger 70B parameter version.

The models possess a context length of 8,000 tokens, giving them competitive abilities when it comes to retaining and understanding user input.

Meta itself has stated that its models are competitive with available closed-source market offerings, and many users are reporting positive initial experiences with Llama 3.

At the same time, AI benchmarking is not an exact science, with many different approaches to evaluating model capabilities, it can be hard to truly construct a hierarchy of AI models. At the same time, Meta’s new model is clearly a success when it comes to user experiences, and its commitment to a largely open-source approach and rapid integration across its products are differentiators.

The State of Open Source AI

Meta’s latest release makes an impressive case for the future of open-source AI. Open-source models are increasingly demonstrating their ability to stand toe-to-toe with the proprietary offerings of the major AI providers. At the same time, it remains unclear whether this gap will close.

In the past, OpenAI has managed to ‘leapfrog’ the state of the art with its releases, which, if it continues to be the case, could mean that open-source is playing a game of perpetual catch-up.

Further, there are questions about whether it is even appropriate to call Meta’s Llama models open-source software.

While the models are released for free public use, they also come with licensing restrictions that do not stand up to the strictest definitions of open-source software. Some call into question the ethics of an open-source movement led by major corporations, rather than decentralized developers or non-profit organizations. Still, the Llama 3 series puts an impressive amount of power in the hands of users.

Bottom Line

Meta remains committed to a largely open-source approach that distinguishes it from other leading AI providers. It remains to be seen whether models  like Llama 3 will become the preferred choice for new AI startups, but Meta has taken a strong step forward with its latest release.

 

---

 

UPCOMING WEBINARS

 

 

Impact of Artificial Intelligence & GPU Computing on Your Architecture

The move toward cloud computing has been underway for years, but nearly all of the profits have gone to a handful of vendors. Artificial intelligence and GPU computing models are beginning to change this trend.

The era of edge computing era is here and with it will come new deployment and business models.

Key topics to be covered in this webinar:

• Artificial intelligence and Computing Trends Driving Modern Architectures
• Architecture Scenario for 2026
• Prepare for these New Architecture Models

 

Register for Free Here

 

 

 

 

The Race to Generative AI: Putting CoPilots to Work in Your Enterprise

Generative AI is here and with it the promise of increased productivity. But is that a promise or is it a reality?

In this webinar Aragon analyst, Adam Pease and CEO, Jim Lundy take on the challenge of putting Generative AI to work.

 

Register for Free Here