Meta Platforms, the parent company of Facebook, has introduced its latest AI model, Llama 3.1, which CEO Mark Zuckerberg describes as “state of the art” and a formidable competitor to offerings from OpenAI and Google’s Alphabet. Developed over several months and with a significant investment of hundreds of millions of dollars in computing power, Llama 3.1 represents a significant update from its predecessor, Llama 3, released in April.
In an interview on Bloomberg’s The Circuit with Emily Chang, Zuckerberg emphasized the importance of intelligence in an AI assistant, stating that the Llama models being developed by Meta are among the most advanced in the world. Work is already underway on Llama 4, indicating the company’s commitment to pushing the boundaries of AI innovation.
Meta executives highlight the broad range of capabilities offered by Llama 3.1, primarily geared towards powering chatbots within Meta’s ecosystem and by external developers. The new model boasts improved reasoning abilities, enabling it to solve complex mathematical problems and synthesize entire texts instantly.
Additionally, Llama 3.1 includes generative AI features that can create images based on text prompts. Notably, the “Imagine Yourself” feature allows users to upload an image of their face, which Llama can then use to generate depictions of them in various scenes and scenarios.
Meta leverages the power of Llama models to drive its AI chatbot, Meta AI, which operates within its apps, such as Instagram and WhatsApp, as well as a standalone web product. With “hundreds of millions” of users already utilizing Meta AI, Zuckerberg confidently predicts that it will become the most widely used chatbot globally by the end of the year.
Moreover, Zuckerberg anticipates that organizations outside of Meta will adopt Llama to train their own AI models, emphasizing its potential as a flexible tool for customization. By providing this technology, Meta aims to empower various entities to create their own models rather than relying solely on off-the-shelf solutions offered by competitors.
Meta’s investments in AI have been substantial, with the Llama 3 models alone requiring “hundreds of millions of dollars” in computing power. Zuckerberg acknowledges that future models will demand even greater investments, potentially reaching billions of dollars in compute power.
Despite the significant investment, Meta is committed to making Llama’s technology accessible to the public for free, provided users adhere to the company’s acceptable use policy. Zuckerberg hopes this open-access approach will establish Meta’s work as the foundation for other successful startups and products, granting the company greater influence over the industry’s direction.
While Meta is committed to openness, the proprietary data sets used to train Llama 3.1 remain a closely guarded secret. Zuckerberg explains that, despite the model’s openness, it has been designed with Meta’s specific needs in mind.
Meta utilizes publicly available user posts from Facebook and Instagram, along with other proprietary data sets licensed from external sources. The company maintains that training Llama on data from its platforms does not provide a unique advantage, as Google and other search engines can access much of the public data.
Meta’s dedication to AI-driven technologies is evident through its extensive investments, including plans to acquire approximately 350,000 Nvidia Corp. H100 GPUs by the end of the year. These state-of-the-art processors are fundamental to training large language models like Llama and OpenAI’s ChatGPT, although they come with a hefty price tag, often exceeding tens of thousands of dollars per unit.
Critics of Meta’s open-source approach to AI express concerns about potential abuse and the prospect of geopolitical rivals leveraging the technology to keep pace with American counterparts. However, Zuckerberg believes that restricting access would hinder innovation and prevent the United States from maintaining its leadership position.
While he acknowledges that the US may not always be years ahead in AI advancements compared to countries like China, even a small lead can compound over time, securing a significant advantage.
In the race for AI supremacy, Meta Platforms is unwavering in its dedication to push the boundaries of technology, with Llama 3.1 representing a major stride forward in the company’s AI capabilities. As the AI landscape continues to evolve, Meta’s investments and open-access strategy position it as a key player in shaping the future of artificial intelligence.
