• by Colin Kirkland , February 24, 2023

“Smaller, more performant models such as LLaMA enable others in the research community who don’t have access to large amounts of infrastructure to study these models, further democratizing access in this important, fast-changing field,” writes Meta.

Meta claims that LLaMA is “desirable” in an industry led by large language models because of the minimal computing power it requires to test new approaches, further research in the field and explore new use cases for AI.

Meta's research-centric approach comes at a time when AI language models are at the center of controversy across the globe.

In the past week, several popular Chinese apps have removed access to ChatGPT and Microsoft has rolled back its updated search browser Bing due to disturbing responses.

Due to its focus on research, LLaMA is unlike leading AI language models such as ChatGPT and Bing.

Meta says it is not releasing the program to the general public, but is granting access to specific groups such as universities, NGOs, and industry labs.

However, like other large language models, LLaMA takes a sequence of words as an input and predicts the next word to generate text. Meta says it trains the model with text from the 20 languages with the most speakers.

In a Facebook post, Meta CEO Mark Zuckerberg said large language models have “shown a lot of promise in generating text, having conversations, summarizing written material” as well as “solving math theorems and predicting protein structure,” adding that the company will contribute to this research.

LLaMA will be released in several different-sized models (7B, 13B, 33B, 65B parameters) under a noncommercial license.

“We believe that the entire AI community — academic researchers, civil society, policymakers, and industry — must work together to develop clear guidelines around responsible AI in general and responsible large language models in particular,” the company wrote in its post. “We look forward to seeing what the community can learn — and eventually build — using LLaMA.”