Meta releases earlier versions of open source AI model Llama 3

Meta Llama 3: Meta Platforms announced its latest advancements in the field of generative AI with the release of Llama 3, their newest large language model, along with a real-time image generator. The innovations, directed at rivalling market leader OpenAI, will be integrated into Meta's virtual assistant, Meta AI, which is positioned as a sophisticated offering in comparison to its competitors.

The updated Meta AI assistant will receive more important placement within Meta's ecosystem, including Facebook, Instagram, WhatsApp, and Messenger apps, as well as a dedicated website. The move positions Meta to directly compete with OpenAI's ChatGPT, backed by Microsoft.

The new Meta AI functionalities, showcased on the standalone website, include generating vacation packing lists, playing trivia games, offering homework assistance, and creating art. Meta's push in generative AI involves major investment in computing infrastructure and the consolidation of research and product teams.

In a bid to democratise AI, Meta has been openly releasing its Llama models for developers to use, potentially disrupting competitors' revenue models. However, concerns have been raised regarding the safety implications of this strategy.

Llama 3 introduces enhanced capabilities, including computer coding and multimodal generation (text and images). Future versions will focus on advanced reasoning and planning abilities. The inclusion of image training data in Llama 3 will also benefit Meta's Ray-Ban smart glasses, allowing the assistant to identify objects and provide relevant information to users.

Meta's partnership with Google to incorporate real-time search results into the assistant's responses expands its capabilities. The update also extends Meta AI's availability to more international markets, although challenges remain in Europe due to stringent privacy regulations.

Mark Zuckerberg highlighted Meta AI's intelligence in comparison to other free models, underlining the performance of Llama 3 across various benchmarks. However, concerns persist regarding the performance gap between free and proprietary models, as well as issues with contextual understanding in previous versions.

Meta addressed some of these concerns by using high-quality data to train Llama 3, resulting in improved contextual understanding. Despite these advancements, challenges remain in the development and deployment of generative AI models, particularly regarding data quality and model performance.

(With Reuters inputs)