Meta, formerly known as Facebook, has unveiled Llama 3, the latest iteration of its cutting-edge open-source large language model (LLM). This groundbreaking release sets new standards in performance, surpassing industry-leading models like GPT-3.5 in real-world applications.
In an official blog post announcing the launch, Meta stated, “With Llama 3, we aimed to develop open models that rival the best proprietary models available today.” The initial release includes 8 billion and 70 billion parameter versions, with larger models exceeding 400 billion+ parameters in development and set for release in the coming months, accompanied by detailed research papers.
The development of Llama 3 has been a meticulous process spanning over two years, involving extensive efforts in curating high-quality training data, scaling distributed training, optimizing model architecture, and refining instruction fine-tuning techniques.
Meta’s 70 billion parameter instruction fine-tuned model has outperformed competitors like GPT-3.5 and Claude in human evaluations across 12 crucial usage scenarios, including coding, reasoning, and creative writing. Additionally, the company’s 8 billion parameter pretrained model has set new benchmarks on popular LLM evaluation tasks.
“We believe these are the best open-source models of their class, period,” Meta asserted.
Meta adopts an “open by default” strategy, promoting an open ecosystem around AI development. Llama 3 will be available across major cloud providers, model hosts, hardware manufacturers, and AI platforms.
Victor Botev, CTO and co-founder of Iris.ai, commended Meta’s approach, stating, “Meta’s Llama 3 model launch aligns with the growing emphasis on responsible AI practices and ethical development.”
Accompanying the release are updated AI safety tools, including Llama Guard for risk classification and Code Shield for filtering insecure code suggestions. Meta emphasizes a system-level approach to responsible AI development, urging developers to implement their input/output filtering.
Meta AI, powered by Llama 3, emerges as the world’s leading AI assistant, accessible via Facebook, Instagram, WhatsApp, Messenger, and the web, offering productivity, learning, creativity, and general queries. Multimodal versions integrating vision capabilities are in development, with an early preview for Meta’s Ray-Ban smart glasses.
While some express skepticism about Meta’s motives, Meta’s release raises the bar for openly available LLMs, following Mistral AI’s recent benchmark-setting release of Mixtral 8x22B.
Meta’s commitment to advancing AI transparency and accessibility through Llama 3 represents a significant stride forward in the evolution of open-source AI models.
#Meta #Llama3 #AI #OpenSource #Technology #Innovation #EthicalAI #Responsibility #MetaAI
Leave a comment