Meta, the company formerly known as Facebook, has made a significant move in the AI space by open-sourcing its large language model LLaMA 2. This decision puts Meta in direct competition with OpenAI’s popular GPT-4, which powers tools like ChatGPT and Microsoft Bing.
The announcement was made during Microsoft’s Inspire event, where Meta highlighted its growing partnership with the tech giant and its support for Azure and Windows. As part of the move, Microsoft also provided more details about the AI tools integrated into its 360 platform and their associated costs.
Meta’s decision to open up LLaMA 2 is aimed at providing businesses, startups, and researchers with access to more AI tools for experimentation as a community. The LLaMA 2 model has been trained on 40% more data compared to its predecessor, LLaMA 1, and is said to outperform other language models like Falcon and MPT in various areas, including reasoning, coding, proficiency, and knowledge tests.
Transparency and safety are key factors in this move. Meta reveals that LLaMA 2 has undergone rigorous safety testing, including red-teaming, which involves generating adversarial prompts to fine-tune the model. The open-sourced LLaMA 2 will be accessible through Microsoft’s Azure platform, and Meta plans to make it available through other providers like AWS and Hugging Face as well.
According to Meta, an open approach is crucial in the development of AI models, particularly in the generative space, where technology is advancing rapidly. By giving access to today’s AI models, developers and researchers can stress test them and identify and address issues swiftly as a community.
While Meta first introduced the LLaMA model in February, it faced an early leak on 4chan. Now, with LLaMA 2 open-sourced, we can expect a surge of new AI-powered tools built upon this model. The previous model already received more than 100,000 requests from researchers, indicating that the open-source LLaMA 2 will have an even broader reach and impact in the AI community.