Meta's large language model and AI assistant are receiving upgrades. The company recently unveiled the first models of Llama 3, available in two sizes: 8B and 70B parameters. These models have been integrated into Meta AI, the company's AI assistant. Meta CEO Mark Zuckerberg expressed confidence in the enhanced intelligence of Meta AI, stating it is now one of the most advanced AI assistants accessible to users.
The latest Llama models boast reduced false refusal rates, improved alignment, increased diversity in responses, and advancements in reasoning, code generation, and instruction. Meta emphasized that Llama 3 represents a significant leap forward in open-source models, rivaling proprietary alternatives in performance and capabilities.
While Llama is marketed as open-source, previous versions required companies with over 700 million monthly active users to seek a license from Meta for usage. Meta's future plans for Llama 3 include making it multilingual and multimodal, enhancing context length, and further improving core capabilities like reasoning and coding.
Meta AI's functionality has expanded to assist users in tasks such as restaurant recommendations, trip planning, and enhancing email professionalism. The Imagine feature now delivers sharper images more rapidly, updating as users type. Meta AI is accessible on various platforms including Facebook, Instagram, WhatsApp, Messenger, and web browsers, with plans to introduce multimodal capabilities to Ray-Ban Meta smart glasses.
Meta is currently developing models exceeding 400B parameters, still in the training phase. Zuckerberg anticipates Meta AI becoming a prominent product, competing with established AI assistants like OpenAI's ChatGPT, Microsoft's Copilot, and Anthropic's Claude.
