According to reports from Siliconangle , Meta is seeking to compete with OpenAI in the field of generative AI and will release a new open source large model whose performance may be comparable to GPT-4.
According to reports, this new model is several times more powerful than Llama 2 and will provide functions such as text and image generation and analysis. The model is currently under development and its final functionality is uncertain.
Meta’s ultimate goal is to accelerate the development of next-generation generative AI models that can generate more human-like expressions.
Meta has long taken an open source approach to exposing its large model products and is known to be one of the largest contributors in the industry. This year alone it has released a large number of AI models and training datasets to the AI community. These include the Code Llama large language model optimized for programming tasks; the SeamlessM4T model that can achieve universal on-demand translation in hundreds of languages; AudioCraft, a generative artificial intelligence model for creating music and sounds; and Voicebox, a speech generation artificial intelligence model. It also launched I-JEPA, a computer vision model that can learn like humans, and FACET, a benchmark dataset designed to help researchers audit computer vision models for bias.
However, its model performance has always been below products such as ChatGPT. It remains to be seen whether Meta's new artificial intelligence large model can catch up with OpenAI. Because their capabilities depend on the amount of training data. It is understood that Meta’s Llama 2 was trained with 70 billion training data parameters. OpenAI has not publicly disclosed how many parameters GPT-4 is based on, but others estimate it could be 20 times larger than Llama, based on about 1.5 trillion parameters.