Meta Presented Llama 3 Neural Network
Today, April 18, the next-generation Llama 3 large language model (LLM) was released with open source code. Versions 8B and 70B are the most powerful of the free AI models for their class and size. The developer claims that it works better than Claude Sonnet, Mistral Medium and GPT-3.5.
Mistral 7B and Gemma 7B can no longer be called modern, while in some tests Llama 3 8B does not show significant superiority over them. However, Meta is much more proud of its more advanced model, the Llama 3 70B, which it ranks alongside other flagship models for generative AI, including the Gemini 1.5 Pro, the most advanced in Google’s Gemini lineup. The Llama 3 70B outperforms the Gemini 1.5 Pro in MMLU, HumanEval and GSM-8K tests, but trails Anthropic’s top-of-the-line Claude 3 Opus, beating only the series’ weakest model, the Sonnet, in five tests: MMLU, GPQA, HumanEval, GSM-8K and MATH . Meta also developed its own set of tests, from writing and coding to summarization and inference, in which Llama 3 70B beat Mistral Medium, OpenAI’s GPT-3.5 and Anthropic’s Claude Sonnet.
The main feature of Llama 3 is that it is open to everyone. Developers, researchers, and simply curious users can create, play, and experiment. In addition, she has become smarter and safer. The language model sets a new standard by demonstrating incredible reasoning skills and an improved ability to follow instructions. Much attention has been paid to the safe and responsible use of AI, so along with Llama 3, new tools are available to users: Llama Guard 2, Code Shield and CyberSec Eval 2.
