Meta Uses Rival’s LLM to Enhance Internal Tool
Meta's internal coding tool relies on both Llama and OpenAI’s GPT-4, demonstrating the benefits of using multiple LLMs.
Meta's internal coding tool, Metamate, uses rival OpenAI's GPT-4 alongside the company’s own LLM (large language model), Llama, to help Meta's developers with coding tasks.
A former Meta employee and a current one (who are both choosing to remain anonymous) revealed to Fortune that Meta has been using GPT-4 to enhance Metamate since at least early 2024, despite CEO Mark Zuckerberg's public promotion of Llama as a leading AI model.
In July, Zuckerberg touted the release of a new version of Llama as “competitive with the most advanced models and leading in some areas”, before boldly predicting that future Llama models would “become the most advanced in the industry”.
However, despite Llama's impressive scale and capabilities, Meta's internal tool, Metamate, still relies on both Llama and GPT-4 to effectively handle a wide range of queries and prompts. It supposedly switches between Llama and GPT-4 depending on the query, allowing Meta to leverage the strengths of both models.
This isn't to say that Llama isn't competitive with other LLMs, such as GPT-4. There are many reasons why those with access to leading AI models may wish to combine forces with other models as well. Some of the many benefits of using multiple LLMs include being able to handle a wider range of tasks more effectively, having the flexibility to choose the best model for each specific query, and reducing the chances of system failure.
One of Fortune’s sources declared that the current version of Metamate was “at least as good as an intern,” agreeing that it was helpful for technical work at Meta, but not capable of complex engineering tasks beyond simple coding.