On Thursday, April 18, Meta Platforms, the parent company of Facebook, Instagram, WhatsApp, and Messenger, revealed the early versions of its latest large language model, Llama 3, along with a real-time image generator. This move signifies Meta’s aggressive push to compete with market leader OpenAI in the field of generative AI technology.
These models are set to be integrated into Meta AI, the virtual assistant touted by Meta as the most advanced among its free-to-use counterparts. Meta AI will receive more prominent placement within Meta’s suite of apps, including Facebook, Instagram, WhatsApp, and Messenger, as well as on a standalone website, positioning it to directly challenge Microsoft-backed OpenAI‘s ChatGPT.
This announcement comes amid Meta’s concerted effort to deploy generative AI products to its vast user base, aiming to rival OpenAI’s dominance in the technology sector. Meta has undergone significant infrastructure upgrades and organizational restructuring to streamline the development and rollout of these AI solutions.
Chris Cox, Meta’s Chief Product Officer, revealed that Llama 3 has been equipped with enhanced computer coding capabilities and trained on both text and image data. However, initially, the model will only generate text. Future versions are expected to incorporate advanced reasoning capabilities and support multimodality, generating both text and images simultaneously.
The overarching objective, according to Cox, is to simplify users’ lives by assisting with various tasks, such as interacting with businesses, composing texts, and planning trips.
Meta also announced a collaboration with Alphabet’s Google to integrate real-time search results into Meta AI’s responses, complementing its existing partnership with Microsoft’s Bing.
Expanding beyond the US market, Meta AI will launch in over a dozen countries, including Australia, Canada, Singapore, Nigeria, and Pakistan. However, expansion into European markets poses challenges due to stricter privacy regulations, including the forthcoming AI Act.
Concerns surrounding the use of generative AI models, such as data privacy and potential misuse by developers, have prompted Meta to release models like Llama 3 for free commercial use. However, this strategy has raised safety concerns among critics.
Meta CEO Mark Zuckerberg highlighted the competitive landscape, emphasizing Meta AI’s intelligence and accessibility. He noted that the largest version of Llama 3 is currently undergoing training with 400 billion parameters, boasting a score of 85 MMLU (Massive Multitask Language Understanding).
Addressing previous issues with Llama 2, Meta emphasized improvements in Llama 3, attributing them to the use of high-quality data and synthetic data to enhance coding and reasoning capabilities.
Cox stated that Meta’s approach to sourcing training data remained consistent, with a focus on quality and relevance.