Meta has introduced its most sophisticated artificial intelligence model to date, Llama 3, hugely advancing its AI capabilities with multilingual proficiency and enhanced performance metrics.
The new model has 405 billion parameters and presents a formidable challenge to paid models from industry giants like OpenAI and Amazon. Llama 3 can converse in eight languages, generate superior computer code and tackle complex math problems more effectively than its predecessors.
CEO Mark Zuckerberg announced that the Meta AI chatbot, powered by these new models, is on track to become the most popular AI assistant by year’s end, already used by hundreds of millions globally. Despite the significant costs, Meta offers its Llama models mostly free to developers.
Meta’s strategy could potentially disrupt the market, as developers may prefer the no-cost Llama models over paid alternatives, challenging the revenue models of rivals. The company has also updated its lighter-weight models with 8 billion and 70 billion parameters, all featuring an expanded "context window" for better handling larger user requests, particularly in coding.
In performance tests, Llama 3 has shown remarkable results, nearly matching and occasionally surpassing the leading models. On the MATH benchmark for advanced math word problems, Llama 3 scored 73.8, close to GPT-4o’s 76.6 and ahead of Claude 3.5 Sonnet’s 71.1. In the MMLU benchmark, covering diverse subjects, it achieved 88.6, just shy of GPT-4o’s 88.7 and surpassing Claude 3.5 Sonnet’s 88.3.
Meta also revealed plans for "multimodal" versions of Llama 3, integrating image, video, and speech capabilities, with early tests indicating competitive performance against Google’s Gemini 1.5 and Anthropic’s Claude 3.5 Sonnet.
(Inputs from Reuters)