LIVE TV
India
Photo: OTV
Recommended Stories
In the competition for AI advancements, two major players, OpenAI and Meta, are pushing the boundaries.
Meta recently unveiled its latest Large Language Model (LLM) called the Meta Llama 3, along with an image generator that updates images in real-time as users type.
What is Llama 3?
It is part of Meta's Large Language Model family, introduced in February 2023. The first version came in four sizes: 7B, 13B, 33B, and 65 billion parameters.
Interestingly, the 13B version of Llama reportedly outperformed OpenAI’s GPT-3, which had 135 billion parameters.
Following this success, Meta released Llama 2 in July of the same year, a notable upgrade. Llama 2 came in 7B, 13B, and 70B parameters.
Now, Meta has introduced Llama 3, the latest version, boasting significant improvements in performance and capabilities.
Llama 3, based on the architecture of Llama 2, is available in two sizes: 8B and 70B parameters. Both sizes include a base model and an instruction-tuned version, enhancing performance in specific tasks.
Currently, Meta has released text-based models in the Llama 3 collection, which are multilingual and support multimodal functions. These models can handle longer contexts, such as coding and reasoning.
Meta plans to integrate Llama 3 into its AI systems, including Facebook. When tested against OpenAI's GPT-4, Llama 3 passed the evaluation while GPT-4 did not.
New GPT-4 passes the magic elevator test pic.twitter.com/LbhLzQIb63
— 7oponaut (@7oponaut) April 10, 2024
Additionally, both AI models easily tackled classic reasoning questions during testing. Many tests have demonstrated the capabilities of both tools.
The Llama 3 70B model has particularly impressed, despite GPT-4 boasting 1.76 trillion parameters. Both models have shown similar results across various online tests.
