Meta has launched Llama 3, the latest iteration of its open-source large language model (LLM), claiming it surpasses previous models like GPT-3.5 in real-world applications.
In a blog post, Meta stated, “With Llama 3, we set out to build the best open models that are on par with the best proprietary models available today.”
Initially, Meta is releasing 8 billion and 70 billion parameter versions of Llama 3, with larger 400 billion+ parameter models in development for future release. The development of Llama 3 took over two years, focusing on high-quality training data, model architecture optimization, and innovative fine-tuning methods.
Llama 3's 70 billion parameter model outperformed other large language models, including GPT-3.5, in human evaluations across 12 key scenarios such as coding, reasoning, and creative writing. The 8 billion parameter model also set new benchmarks on popular language model evaluation tasks.
Meta emphasizes an "open by default" approach, making Llama 3 available across major cloud providers, model hosts, hardware manufacturers, and AI platforms.
Victor Botev, CTO and co-founder of Iris.ai, commented on Meta’s commitment to transparency and responsible AI practices. He said, “Meta aligns with the growing emphasis on responsible AI practices and ethical development by embracing transparency through open-sourcing.”
Alongside Llama 3, Meta introduced updated AI safety tools, including Llama Guard for risk classification, CyberSec Eval for assessing misuse, and Code Shield to filter insecure code suggestions.
Botev added, “While open initiatives like Llama 3 promote scrutiny and collaboration, addressing AI challenges requires comprehensive approaches to issues like data privacy, algorithmic bias, and societal impacts.”
Meta's AI assistant, Meta AI, powered by Llama 3, is now integrated into platforms like Facebook, Instagram, WhatsApp, Messenger, and the web, offering users enhanced productivity, learning, creativity, and general queries. Multimodal versions with vision capabilities are also in development for Meta’s Ray-Ban smart glasses.
Despite the advancements of Llama 3, some experts remain skeptical of Meta’s motivations behind the open-source approach. Nonetheless, with competitors like Mistral AI setting new benchmarks, Meta's Llama 3 continues to raise the standard for open-source language models.
