Will Meta’s Llama 3.1 revolutionize the future of open-source AI?
Authored By: Nazeerul Hoque — Writer, NexStudent Network
Edited By: Mohit Prathipati, Cristina Martinez — NexStudent Network
Llama 3.1 — a language model that was released earlier this year — is transforming the way users think about artificial intelligence. The unique feature of this language model stands out against all the competitors out there. Llama offers real-time chats with the power of AI, generates quality images, animations, and much more!
What is Llama?
Large Language Model Meta AI (Llama) is a large language model, or LLM, developed by Meta. Llama aims to compete with leading models such as OpenAI’s GPT-4 Omni, Anthropic’s Claude 3.5 Sonnet, and Google’s Gemini 1.5 Pro. Llama 3.1 comes in three flavors: 8B, 70B, and 405B, each trained on different numbers of parameters. In a series of benchmarks, 405B was on par with leading closed-source models, even surpassing them in certain metrics. The lightweight, less expensive 8B surpassed OpenAI’s GPT-4o mini in. With over 65 billion parameters embedded into the model, it is one of the biggest AI models.
Open Source
Unlike many of its competitors, Llama is an open source, meaning that anyone can host it on their own hardware or modify it to suit their own needs. This ultimately reduces costs for developers, researchers, and businesses while simultaneously reducing their dependence on closed-source models. Many closed-source models require them to be hosted on their servers, which can pose a potential privacy risk for businesses and governmental organizations handling confidential information. By hosting Llama locally, the privacy risk is minimized.
However, critics have pointed out that while the model is open-source, the vast set of training data that was used is not likely because Meta does not have permission to license it. Under Llama’s license, users must explicitly state their product is “Built with Llama.” Nevertheless, the freedom granted by Llama’s license will provide many opportunities for developers and businesses that they did not previously have access to with GPT-4.
Other Open Source Models
Llama is not the only open-source LLM. In June, Nvidia announced Nematron-4 340B, an open-source LLM, notably capable of high-quality synthetic data generation, which could have huge affects on the discovery of new medicinal treatments. The NeMo framework offers many customization and fine-tuning options, making it easily tailored to meet many needs. Mistral, a French startup, released Large 2 a day after the announcement of Llama 3.1. Mistral claims Large 2 generates concise responses, in contrast to many models’ tendencies to be verbose, has a deeper understanding of human culture and produces fewer hallucinations than Large 1. However, ML2 requires a commercial license and is only truly open source for non-commercial and research endeavors.
Developers
Mark Zuckerberg emphasized the open-source nature of Llama in a letter, where he notes that the path for Llama to become the industry standard is by consistently being competitive, efficient, and open generation after generation. Public access enables a large community to continue developing the Llama ecosystem, giving Meta an edge over competitors. Open-source software has historically been more secure, which could minimize the risk of hallucinations and unintentional harm in the future. As AI models from a plethora of companies continue to advance, open-sourcing the best model, at any given moment, won’t significantly disadvantage a company.
Companies have been actively looking for groups of developers to expand their networks and establish their presence in the fast-growing and competitive field of large language models. For example, on May 14, Google launched the Gemini API Developer Competition in an effort to attract developers to their new platform, with a total prize pool of over $1 million USD and a grand prize of a custom electric 1981 DeLorean.
A report by the National Telecommunications and Information Administration, part of the US Department of Commerce, recommended the government to “incentivize global and domestic research and innovation that harnesses the many benefits of open foundation models.” The report defines open foundation models as AI models that are trained on an extensive set of training data, contain tens of billions of parameters, and have their model weights widely available to the public.
Please follow NexStudent Network for more tech news!
More about our organization. We are a non-profit organization teaching free first-rate education about programming in Python. Over 750 students across the nation are enrolled in our program already with tons of positive reviews!
Join us — links below!
