When it comes to open source AI, there is no doubt that Meta has been leading the way. For nearly a decade, Meta AI has been a pioneer in open-sourcing high-impact AI-related tools and technologies, including notable contributions such as PyTorch, React, and the original LLaMa model. Now, with Llama 2, the company is taking further steps to bolster its open-source stance.

This approach has provided Meta with widespread adoption of their technology, with developers trying out the model in every way possible, just like PyTorch. But for Meta, the benefits still remain to be seen.

Interestingly, a lot of companies have already started to adopt Llama 2 within their systems. For example, Perplexity AI, LunaAI, Poe, and others have integrated Llama 2 into their chatbots. And many other smaller companies have started leveraging Llama 2 with their proprietary data through platforms like DataBricks and Amazon SageMaker. Siding with the open source community, Meta seems to be playing the good guy by allowing others to use something they built very dearly to be used commercially by others. How exactly does Meta profit with this?

The Meta Business Strategy is Risky

If we take example of Threads, the Twitter alternative that the company released recently, it has been so dead within a month of its launch that the company is trying to introduce AI personas to keep the users from continuing using it. Same is the case with the Metaverse, where more than Meta, people have been thinking about Apple these days with the announcement of the Vision Pro headset. In AI, the company is hell bent on siding with open source instead of its customer facing products.

This is probably because the company’s PyTorch’s open-sourced nature helped it gain widespread adoption, making it one of the most popular deep learning frameworks in the world. Same is the case with React. Llama 2’s open-source release aims to continue this legacy and advance the boundaries of AI research while fostering collaboration and further innovation. 

Every new development in PyTorch by Meta has been driven by a non-profit approach by the company. For example, with the introduction of the PyTorch Foundation for accessing SOTA AI models, the company decided to do it openly on Linux for transparency. The GitHub repository already boasts more than 150,000 projects built on it. Similarly, the number of projects on Llama 2 is also increasing on Hugging Face.

On the other hand, to make profit out of putting Llama 2 out in the public and keeping it restricted for big tech, Mark Zuckerberg said, “the largest companies with public cloud offerings don’t just get a free licence to use this. They will need to make a business arrangement with us”. This seems to be the way forward for Meta with Llama 2, considering Zuckerberg also thinks that Meta is four years behind OpenAI. 

Regardless, Yann LeCun’s endorsement of Llama 2 only reinforces the potential impact that this open-source language model can have on the AI community. By providing free access to Llama 2’s capabilities, Meta is enabling startups and smaller organisations to leverage state-of-the-art AI technology, levelling the playing field and empowering the next generation of AI-driven applications. Meta hopes that it will be able to fix a lot of problems with the current LLM models, drive research, while also making a few bucks for itself, while doing everything transparently. 

Meta probably does not care about revenue from generative AI

Meta’s biggest source of revenue is its advertising business. With its simple integration of AI into all its apps, it might be able to make the best out of AI right now. The company also reported its most profitable quarter after 2021. On the other hand, companies like Google and OpenAI can keep progressing towards building their models better.

The success of PyTorch serves as a compelling example of how Meta’s open-source strategy can lead to significant achievements. As an open-source deep learning framework, PyTorch gained rapid popularity within the AI community. Its user-friendly interface, dynamic computation graph, and strong community support made it a favourite among researchers and developers alike. Consequently, PyTorch became the go-to platform for numerous AI projects and research initiatives, cementing its position as one of the leading frameworks in the field.

Another avenue is leveraging the open-source model to attract top talent and foster innovation within Meta itself. Access to Llama 2’s codebase can entice AI experts to work with Meta, further strengthening the company’s AI research and development efforts. Additionally, the adoption of Llama 2 by external developers can lead to broader integration into various products and applications, potentially increasing Meta’s visibility and influence in the AI ecosystem. That is all it wants at the moment.

Conclusively, building on the success of PyTorch, Llama 2 is poised to become a driving force behind groundbreaking AI projects, inspiring the next generation of researchers, developers, and entrepreneurs. As Meta continues to embrace the open-source approach, it sets the stage for a more inclusive and thriving AI community, where accessibility, collaboration, and shared knowledge drive the future of AI.