Meta, the Facebook parent, is building a new AI model aimed at competing with GPT-4, OpenAI’s most advanced model, Wall Street Journal reported on Sunday citing unnamed sources. The company is aiming for a release of the model next year, which, according to the report, will be significantly more powerful than the recently released Llama 2, an open-source AI model that Meta launched earlier this year.
The social media company has been making substantial investments in generative AI and has recently released several other models, in the last few months, including SeamlessM4T AI that can transcribe and translate up to 100 languages, and CM3leon, an image generation model.
The new model by Meta will be aimed at businesses and would help them produce text and analysis so it would apparently compete with ChatGPT’s recently launched enterprise version and Coral by Cohere.
The report noted that Meta expects to begin training the new model early next year. Meta has been buying Nvidia’s H100 chips in large quantities and is projected to have the second-largest number of H100 chips in the world by next year, according to analysis by SemiAnalysis. These chips are used to train large language models.
Mark Zuckerberg, the CEO of Meta, has been pushing to make the latest model open-source, according to Wall Street Journal.
Meta is not the only big technology company trying to ramp up the efforts to compete with the likes of OpenAI and Microsoft. Apple is reportedly also building an AI chatbot which is internally being referred to as Apple GPT, and is based on company’s own large language model Ajax.