Revolutionizing the AI Industry: Meta’s Llama 3 Launches with Improved Reasoning Capabilities

In recent news, Meta has launched the latest version of its open-source large language model, Llama 3. The company has introduced models trained with 8B and 70B parameters, which are expected to revolutionize the AI industry. These models have been designed to support a wide range of use cases with improved reasoning capabilities, making them the best-in-class open source models available in the market.

The initial release of the models was initially anticipated in May, but Meta decided to launch the first two models of the new Llama 3 generation earlier. These text-based models have been trained and fine-tuned with 8 billion and 70 billion parameters, showcasing advancements in reasoning, code generation, and instruction capabilities. They offer state-of-the-art performance across industry benchmarks and represent a significant improvement over the previous Llama 2 model.

A comparative analysis with other models of similar parameter scales has shown that Llama 3 outperformed in the MMLU benchmark, demonstrating superior overall knowledge. Meta has conducted evaluations to highlight the model’s capabilities in various use cases like advice-giving, brainstorming, topic classification, and coding skills among others.

To train these powerful models, Meta used a large dataset of more than 15T tokens from publicly available sources. This dataset is seven times larger than the one used for Llama 2. The training data includes information in over thirty languages covering a broader linguistic scope. Additionally, Meta has emphasized its commitment to responsible development by introducing resources like Llama Guard 2, Code Shield, and CyberSecEval 2 to ensure safe usage of these powerful tools.

These new models have been integrated into Meta’s AI assistant on various platforms like Facebook, Instagram, WhatsApp