In the race for AI supremacy, companies like Microsoft, Google, and Meta have been developing and launching their own large language models (LLMs). These models, such as OpenAI’s GPT-4, Google’s Bard, and Meta’s LLaMA, have generated much excitement and speculation about which one will come out on top. However, a recent development suggests that the future of AI may not be a competition between these models, but rather a collaboration.
Meta, the company behind Facebook’s rebranding, made headlines when it announced that it would offer its LLaMA model for free under an open license and bring it to Microsoft’s Azure platform. This move demonstrated the value of interoperability in the AI space, and it is likely just the beginning of more partnerships to come.
Traditionally, LLMs have been siloed and offered within controlled environments where users required permission to use the model or access the training data. For example, OpenAI releases new versions of its GPT model and provides developers with paid API access. Apple is also developing its own LLM called Ajax, which is not yet publicly available. Google’s Bard is not open source either.
However, Meta’s LLaMA was always intended to be open source and democratize access to AI. While initially only accessible through Meta, the company has now made it available to Azure users and partially removed the licensing fee requirement. This move not only aligns with Meta’s commitment to open development but also makes business sense by expanding the user base for LLaMA.
Collaboration between competitive tech companies is not uncommon. Meta has previously partnered with Microsoft to bring its Teams product to Workplace by Meta, which runs the Office 365 suite. This willingness to work together demonstrates that even in a fiercely competitive market, companies recognize the benefits of cooperation.
However, openness and interoperability also come with risks. Ilya Sutskever, co-founder and chief scientist of OpenAI, has expressed regret about sharing research due to concerns about competition and safety. Opening up datasets can also increase the risk of copyright infringement lawsuits if the sources used for training the models are revealed.
Despite these risks, advocates of AI interoperability argue that it is necessary for the growth and evolution of AI. By allowing AI systems to work together, developers can achieve better results, provide more comprehensive services, and avoid overreliance on a single source of information. Additionally, having a variety of LLM frameworks to choose from can lead to cost and time savings in development.
While Meta’s decision to bring LLaMA to Azure is a positive step towards open source and interoperability, there are still challenges to overcome. Currently, there is no bridge that allows communication between apps built with LLaMA and those running on OpenAI’s GPT models. Furthermore, there are debates over whether LLaMA meets all the criteria for open-source software, as it does not use a license approved by the Open Source Initiative and restricts commercial use without a fee for developers with over 700 million monthly active users.
Overall, the availability of LLaMA on Azure is a move in the right direction for open source and interoperability. It allows developers easier access to different LLM models and sets the stage for future collaboration. While healthy competition will continue to drive innovation in the AI industry, the true advancement of AI lies in the ability of companies to work together and create an interconnected ecosystem.
