A new AI language model published Tuesday by Meta: Llama 3.1 405B is causing excitement in the field. The rationale is: This is maybe the first time anybody could download and run a free GPT-4-class large language model (LLM) on their own hardware. You still need some hefty hardware. Meta claims it runs on a “single server node,” which isn’t desktop PC-grade gear. Still, it’s a fascinating view across the bow of “closed” AI model providers like OpenAI and Anthropic.
Meta claims, “Llama 3.1 405B is the first openly available model that rivals the top AI models when it comes state-of- the-modern capabilities in general knowledge, steerability, math, tool use, and multilingual translating.” Mark Zuckerberg, the CEO of Company, refers to 405B as “the first frontier-level open source AI model”.
“Frontier model” is a phrase used in the artificial intelligence sector to describe an AI system meant to challenge present capabilities. Meta is therefore establishing 405B alongside the leading AI models in the field, like Google Gemini 1.5 Pro, OpenAI’s GPT-4o, and Claude’s 3.5 Sonnet.
Under benchmarks like MMLU (undergraduate level knowledge), GSM8K (grade school arithmetic), and HumanEval (coding), a Meta chart shows 405B comes quite near to matching the performance of GPT-4 Turbo, GPT-4o, and Claude 3.5 Sonnet.
source
date: 2024-07-27 12:00:00
duration: 00:07:58
author: UCPORnAp6o9u0Cr8gGNA06iQ
