Meta introduces Llama 3.1: What is an open-source AI model? How can it change the world?

Such models allow developers to modify them to revise their recommendations and decisions to adapt to their specific needs. These modified versions can later be shared with the public for any purpose.

In the last three years, there has been a significant rise in the number of enterprises and independent developers using open source AI technologies.

Meta CEO, Mark Zuckerberg attributes this rise to organisations seeking independence from closed vendors to avoid changes such as model alterations and service disruptions. They prefer open-source models as they come with flexibility and control.

According to Zuckerberg, organisations need to train, fine-tune, and distil their own models to meet their specific requirements while using different-sized models for various tasks. While small models are suited for on-device and classification tasks, larger models are adept at managing complex tasks.

The Meta CEO makes the case for investing in open-source ecosystems, seeing it as advantageous for long-term development and faster progress. The transparency and security of open source technology enable models to run securely on any infrastructure.

What do we know about Llama 3.1?

Llama 3.1 is Meta’s largest ever open source AI model, and the company claims that it has outperformed the likes of OpenAI’s GPT-4o and Anthropic’s Claude 3.5 Sonnet on numerous benchmarks.

The new Llama 3.1 models are reportedly far more complex compared to Llama 3 models.

In his blog post, Zuckerberg predicted that Meta AI will be the most widely used AI assistant by the end of this year, surpassing ChatGPT. This could be possible once Meta rolls out the Llama-based Meta AI assistant in more countries and languages.

“We’re releasing Llama 3.1 405B, the first frontier-level open source AI model, as well as the new and improved Llama 3.1 70B and 8B models,” the Meta CEO said. “In addition to having significantly better cost/performance relative to closed models, the fact that the 405B model is open will make it the best choice for fine-tuning and distilling smaller models.”

Zuckerberg claimed that running inference (a process to train machine learning models) on Llama 3.1 405B is cost-effective as it is around 50 per cent cheaper than closed models like OpenAI’s GPT-4o.

Additionally, Llama comes with systems like Llama Guard that can be secure against unintentional harms such as bad health advice or unintended self-replication, which he noted as major concerns.

Why is Meta batting for open source AI?

According to Zuckerberg, Meta’s prime focus is building the best experiences and services for people by giving them access to the best technologies and allowing for better product development without being limited by closed ecosystems offered by competitors.

The open sourcing of Llama, according to him, encourages an ecosystem of tools and improvements that can likely ensure long-term access to the best technologies.

Another feature is the cost-effectiveness of the open source model. While the open source nature of the Llama model does not affect Meta’s bottomline or investments in research, Zuckerberg highlighted Meta’s history of successful open-source projects such as the Open Compute Project and tools like PyTorch and React, as they saved billions and drove innovation for Meta.

The Meta CEO believes that open source AI is safer owing to its transparency and wide scrutiny. It allows larger organisations to counteract threats from smaller ones, promoting security and stability, as open source AI models address intentional harm by bad actors.

Zuckerberg makes the case for investing in open source AI to boost future innovations and benefit startups, universities, and nations with scarce resources, leading to economic opportunities for all. Its applications would enhance human productivity and creativity and improve our quality of life at the same time, accelerating economic growth and scientific research.