Meta’s Llama 3: Developing Artificial General Intelligence
Meta CEO Mark Zuckerberg announced on Thursday (18th January 2024) that the tech giant has officially started training Llama 3 - the next generation of its generative AI (Gen AI) model - to develop artificial general intelligence (AGI)
The company joins other high-profile developers like OpenAI and Google DeepMind also exploring the same type of AI. Although still hypothetical, AGI could become the most powerful technology ever invented.
AGI is a type of AI that could be powered by human-level intelligence and has prompted much anxiety over if it could surpass humanity altogether.
Open Source: Making AI technology accessible for all
“Our long term vision is to build general intelligence, open source it responsibly and make it widely available so that everyone can benefit,” says Mark Zuckerberg via an Instagram post. “We’re bringing our two major AI research efforts (FAIR and GenAI) closer together to support this.”
According to Zuckerberg, the company’s research group FAIR will be moved to the same part of the company as the Gen AI products team. Meta’s goal is to reach its billions of global users more directly via the power of AI.
With the company now developing Llama 3, the plan is for it to have a wider range of capabilities. This includes coding, much like Google’s new Gemini AI model. Zuckerberg highlights that these future models will continue to be trained responsibly and safely, as Meta continues to commit to open sourcing its AI.
Llama 3 is a proposed Gen AI text model that is believed to rival the gold-standard: OpenAI’s GPT-4. Meta’s most recent model, Llama 2, is openly available.
Open sourcing AI in this way not only provides developers with greater accessibility to harness their own systems, but also promotes the importance of transparency within the wider technology industry.
When the base Llama model was first introduced in February 2023 (Llama 1), less than one year ago, it allowed researchers to study them without having substantial infrastructure in place. It was hailed for working to democratise access to the rapidly developing field of AI, having open sourced its model.
With Llama 2 subsequently launched in July 2023, Meta announced that its capabilities were improved, having been pre-trained on two trillion tokens. This therefore doubled the context length to Llama 1, with its fine-tuned models also trained on more than one million human annotations.
Developing AGI for deeper digital understandings
The release of Llama 2 ultimately prompted Meta to consider AGI as a way of developing smarter AI for its customers. Mark Zuckerberg has said that the company’s recent decisions are designed to help build AGI systems that will give its products more human-like capabilities in the near future.
“It’s become clearer that the next generation of services requires building full general intelligence,” Zuckerberg says in his announcement.
“Building the best AI assistants, AIs for creators, AIs for business and more—that needs advances in every area of AI, from reasoning to planning to coding to memory and other cognitive abilities.”
With AGI being posited as having human-level intelligence, it could solve a current barrier with AI in its ability to generalise knowledge. Harnessing AGI could lead to deeper understanding and problem-solving abilities, as systems would not only be able to adapt to a wide range of situations, but also engage in critical thinking.
However, whilst these announcements are exciting, it is important to acknowledge the ethical implications of creating a technology at the same intelligence level as a human. Meta continuing to open source its developments and displaying a commitment to AI ethics will ultimately work to shape a positive future for human and AI coexistence.
******
Make sure you check out the latest edition of AI Magazine and also sign up to our global conference series - Tech & AI LIVE 2024
******
AI Magazine is a BizClik brand
