Meta is taking competition head on in every field. It started competing with Elon Musk’s X and launched Threads. To compete with OpenAI’s ChatGPT, it launched Llama, and then Llama 2. Threads is a flop, but when it comes to competing with OpenAI, Meta has it surrounded with its tiny llamas from every direction. It even partnered with Microsoft to host its open source and commercially available LLM on the cloud to compete with OpenAI’s GPT.
Now, Meta has another trick up its sleeve.
Meta is planning to launch a platform to help developers generate code automatically. According to a report by the Information, Meta’s code generating platform, which is dubbed as Code Llama, will be based on Llama 2 and is going to be open source. It is expected to launch as soon as next week, as told by someone familiar with the matter. Meanwhile, DeepMind’s AlphaCode, its own code generating platform, has been available on GitHub since last year. But arguably, it has barely made any impact.
Moreover, this is something that OpenAI has been doing all this while through Codex, or Amazon’s CodeWhisperer. But interestingly, Both of these are proprietary and closed source, something that “OpenAI” has been known for now for some time. But when it comes to Meta, it is building upon its biggest moat – the open source community.
What’s in this for Meta?
Though Microsoft has been trying to make everyone a developer by providing GitHub Copilot and even embedding some coding capabilities on ChatGPT, Meta is on a roll. Through Code Llama, companies and enterprises would be able to make their own AI apps, without paying for OpenAI’s or Google’s paid software.
Just like Llama 2 was allowing people to build their own chatbots, Code Llama will enable companies to build their own AI coding assistants. The best part for enterprises here is that since it is open source, developers from companies would be able to upload their source code and generate a lot of specific code based on the uploaded proprietary code.
This has been a continuous worry that has been pushing enterprises away from using OpenAI services, something that even Microsoft has acknowledged. That is why a lot of people have been banning the use of ChatGPT for their employees.
For Meta, there might be no direct monetary benefits of open sourcing its software. Meta revealed in May that all this while, the company has been using CodeCompose internally to generate code with AI. There is no proof that Code Llama is going to be an extension of it but the company has said that the model was trained on legally available programming data, unlike competitors like OpenAI’s Codex.
Furthermore, as Meta continues to open source its models, it would benefit from the contributions from the community and build on its products. For instance, the company has been developing chatbot personas for its social media platforms with the help of Llama 2. Developers all over the world have been able to spot the security flaws of the model and helped Meta improve it over time. Same could be the case for Code Llama.
Code Llama to bring a revolution
There has been this talk that open source is actually the moat for Meta, which OpenAI and Google lack. This is being proven again and again ever since a lot of companies have started adopting Llama 2 to build their own AI products, and moving over OpenAI and Microsoft proprietary algorithms. But lately, things haven’t been looking great for OpenAI amidst all of these open source models.
According to reports, OpenAI has also been working on an open source model to compete with Meta’s success. Code named G3PO, OpenAI has been in talks to work on this open source model. But clear information on the release date or if it would even go ahead and release it is unsure.
The only thing that Meta is still struggling with is that Llama 2 still requires a significant amount of computer power. Same would be the case with Code Llama. For this, there have been several solutions coming up. One such example is Petals, which is leveraging a decentralised pipeline, much like sharing compute on crypto mining or torrent, to increase the inference speed of neural networks.
Nonetheless, Meta’s Code Llama is going to increase the pressure that OpenAI has already been going through lately. Possibly, it might even push the company to release an open source model to not lose on its GPT and Codex customers because of privacy issues.