The possibilities have expanded significantly: Meta has released a new AI

Meta, the tech company that owns Facebook and Instagram, has unveiled a new model of its proprietary artificial intelligence called Llama 3.1 405B. Its creators called it "the next generation of the most advanced large-scale open-source language model."

In a message published on the AI website, Meta claims that the development of their specialists is the largest and most powerful basic model in the world available to all users. The creators claim that Llama 3.1 405B can compete with OpenAI and Anthropic in general knowledge, math skills, and language translation. The index in its name indicates that the model has 405 billion parameters.

According to Meta executives, as a result of training on the fastest chips available on the market, Llama 3.1 405B was able to outperform competitors in 150 tests. The developers of other popular artificial intelligence do not disclose the key parameters of their models, but it is known that OpenAI's GPT-4 has about 1.5 trillion parameters. Along with the release of Llama 3.1 405B, Meta also introduced updated versions of other Llama models that contain 70 billion and 8 billion parameters, respectively.

Meta said that with Llama 3.1, it wants to allow everyone to make the most of the 405B model's capabilities. Among the features provided to users:

• real-time and batch output;
• controlled fine-tuning;
• evaluation of your model for a specific application;
• continuous pre-training;
• search-aided generation (RAG);
• function calling;
• generation of synthetic data.

All these advanced features are available to users immediately. Developers can also explore advanced workflows, such as easy use of synthetic data, follow pre-built model distillation instructions, and ensure seamless RAG with solutions from partners including AWS, NVIDIA, and Databricks. In addition, Groq has optimized low-latency output for cloud deployments, and Dell has achieved similar optimization for on-premises systems.

Llama 3.1 405B improvements include enhanced inference capabilities, broader multilingual support, and a significant increase in context length to 128K tokens. Meta said that the presented model will be used in Meta AI, the company's chatbot that works on various platforms, including Facebook, Messenger, Instagram, and WhatsApp, as well as on smartphones.

Mark Zuckerberg, the head of Meta, also spoke about the presented AI in his blog. He emphasized that artificial intelligence should remain accessible to everyone as an open-source tool. However, the company did not elaborate on the availability and deployment of the model. Currently, Llama 3.1 is delivered via the Meta AI service. To use it, you need to log in with Facebook or Instagram. The service is not yet available in Ukraine.

Subscribe to OBOZ.UA channels on Telegram and Viber to keep up with the latest developments.