Meta recently made an announcement that it is open-sourcing its large language model called LLaMA 2. This move allows for its free use in both commercial and research applications, putting it in direct competition with OpenAI’s GPT-4, which is also available for free and powers tools like ChatGPT and Microsoft Bing.
The announcement was made by Meta during Microsoft’s Inspire event, where they emphasized their growing partnership with Microsoft. They highlighted that LLaMA 2 will be supported on Azure and Windows platforms. In addition, Microsoft shared more details about the AI tools integrated into its 360 platform and the cost associated with them. This partnership extends further, as Qualcomm also revealed its collaboration with Meta to bring LLaMA to various devices such as laptops, phones, and headsets, starting in 2024. This endeavor aims to enable AI-powered applications that can function without relying on cloud services.
According to Meta’s press release, the decision to open-source LLaMA is driven by the objective of providing businesses, startups, and researchers with access to more AI tools. This move enables experimentation and collaboration within the AI community. LLaMA 2 was trained on 40% more data compared to its predecessor, LLaMA 1, and includes information gathered from publicly available online sources. Meta claims that LLaMA 2 outperforms other large language models like Falcon and MPT in various areas such as reasoning, coding, proficiency, and knowledge tests.
In a statement, Meta expressed their belief that an open approach is essential for the development of AI models, particularly within the generative space where technology is rapidly advancing. By opening access to AI models, they anticipate that developers and researchers will be able to identify and solve problems quickly in a collaborative manner.
LLaMA 2 will be available through Microsoft’s Azure platform, but Meta intends to make it accessible through other providers such as AWS and Hugging Face. This broader availability aims to ensure that users can leverage LLaMA across multiple platforms.
Back in February, Meta initially announced the LLaMA model. However, it quickly leaked on 4chan a few days later, prompting Meta to take steps to address safety and transparency concerns. They describe how LLaMA 2 has undergone rigorous testing for safety, including internal and external evaluation through the generation of adversarial prompts to facilitate model fine-tuning. Meta is committed to ensuring that the models are continuously evaluated and modified to improve their safety and effectiveness.
In essence, Meta’s decision to open-source LLaMA 2 demonstrates their commitment to advancing AI research and development by fostering collaboration and experimentation within the community. With the availability of LLaMA 2 on various platforms, businesses, startups, and researchers will have broader access to powerful AI tools, allowing them to explore new possibilities and drive innovation in their respective fields.
