We built our large language AI models (LLMs) to not only create value in Meta’s apps and technologies, but to also allow other organizations to benefit from these models by open-sourcing them. Today, Meta Llama, our collection of open-source large language models are already being used by organizations in education, customer service, research and medicine. Our Llama models have more than 170 million downloads. Here are a few examples that showcase what’s possible with Llama 2 and 3.
Using Llama to Create a Localized Math Education Platform
STEM education is a critical factor in economic mobility and innovation. With this in mind, Mathpresso, a South Korean education tech startup, along with QANDA, which offers personalized math learning content in 50 countries, used our Llama 2 model to create MathGPT, an LLM with powerful, accurate math capabilities.
”Commercial LLMs like ChatGPT lack the customization needed for the complex education landscape,” said Mathpresso co-founder Jake Yongjae Lee.
According to Lee, student learning is influenced by hyperlocal factors including curriculum, school district, exam trends, teaching style and more. Llama 2 is navigating that complexity, and helping Mathpresso create a world where more people have access to quality education. Using Llama 2, the team was able to create flexible, localized educational products while utilizing their own specialized data and techniques.
Earlier this year, MathGPT set a new world record in benchmarks assessing mathematical performance at both the primary and secondary school level.
An AI Companion for Zoom Workplace and Zoom Business Services
Zoom leveraged Llama 2 and other third party models to create an LLM that powers their generative AI assistant, Zoom AI Companion. AI Companion can catch you up on what happened in a meeting if you need to step away from your desk. It can also synthesize Zoom Meetings, highlight next steps and takeaways, and provide helpful presentation tips for your next meeting. AI Companion can also summarize your unread messages in Zoom Team Chat and help you craft responses that fit the length and tone you’re looking for.
Adapting Llama to Provide Medical Information in Low-Resource Settings
In medicine, access to the right information at the right time can determine everything.
Researchers at EPFL’s School of Computer and Communication Sciences and Yale School of Medicine adapted Llama 2 to compress medical information into an accessible, conversational interface. The result is Meditron, a suite of large multimodal foundation models designed to assist with clinical decision-making and diagnosis. In working closely with humanitarian organizations like the International Committee of the Red Cross (ICRC), Meditron has the potential to assist healthcare workers in diagnosing and treating patients in underserved areas, and serve emergency scenarios that require fast and accurate medical response.
Following our recent release of Meta Llama 3, the team fine-tuned the new model within 24 hours to deliver Llama-3[8B]-MeditronV1.0. Meditron achieves strong results on leading industry benchmarks in the field, such as question-answering of biomedical exams.
“Foundation models have become modern-day intellectual and cultural assets,” says Yale professor Mary-Anne Hartley, who is co-leading the project. “When applied to the medical domain, they have the potential to provide life-saving advice and guidance. Yet the lowest-resource settings have the most to gain and remain the least represented.”
The goal is that open models like Meditron can help create equitable access to medical knowledge.
These are just a few examples of how organizations are using Meta Llama to reach their goals and make an impact on people’s lives. We’ll continue to share how others put Meta Llama to good use.