Despite Mark Zuckerberg praising Meta’s Llama AI model as one of the best in the technology, his company is happy to use a rival when necessary.
Meta’s internal coding tool, Metamate, incorporates OpenAI’s GPT-4 model, along with Meta’s own Llama model, to make the tool more useful for developers and other Meta employees who use it, two people who have used the tool said. Wealth. OpenAI’s model has been part of Metamate since at least the beginning of this year.
The people, a current and former Meta employee who requested anonymity for fear of retaliation from Meta, noted that they found Metamate useful. It appears to pull answers to coding questions or queries from either Llama or GPT-4, they said, depending on the type of query or how the employee uses the tool.
“It’s one of the better genAI things they’ve done,” one of the people who used the tool said of Meta’s efforts in generative artificial intelligence. Metamate is similar to other AI coding tools like Microsoft’s GitHub CoPilot or Anthropic’s Claude Sonnet.
Separately, the Chan Zuckerberg Initiative, a large philanthropic organization run by Zuckerberg and his wife Priscilla Chan, is also an OpenAI customer, two people familiar with the entity said. Wealth. CZI is developing an “educational” genAI tool based on ChatGPT with some additional customizations or what is often called a “wrapper” of the genAI technology. Earlier this year, Sam Altman, CEO of OpenAI, was appointed as the new CEO AI Advisory Board in CZI.
Meta’s spokesman declined to comment. CZI representatives did not respond to emails requesting comment.
While it’s not uncommon for a company implementing genAI tools for its workforce to rely on more than one model, Zuckerberg positioned Meta as a key player in the model wars. It’s a battle most companies are waging against OpenAI, which released ChatGPT two years ago and began a race to catch up to gain a foothold in what is widely considered a new era of technology. Zuckerberg has spent the past year touting Llama as an “open source” alternative that’s as good or better than fully closed models, such as those from OpenAI and Google.
When Meta released a new version of Llama mid-year, said the CEO it was “competitive with the most advanced models and leading in some areas.” Zuckerberg stated that by next year, the Llama model will be “the most advanced in the industry.”
Even so, a tool like Metamate and its underlying Llama model still needs the help of another like GPT-4 to function well. And it doesn’t matter that Llama is one of the biggest models in the world, with the latest version pre-trained on trillions of tokens, individual bits of information fed into LLM, and four times the computer code of its predecessor.
One of the sources he talked to Wealth agreed that the current version of Metamate, originally called Code Compose, is useful in technical work at Meta, noting that it is “at least as good as an intern.” Translation: It’s good at basic coding, but not for more complex engineering work.
However, a side effect of AI assistance is that top executives think they need fewer workers. “That’s part of what’s driving the layoffs,” the person added.
Indeed, Meta has to release gradually a dozen workers this year after laying off more than 20,000 employees between 2022 and 2023.
Are you a Meta employee or someone with insight or advice to share? Contact Kali Hays securely Signal at +1-949-280-0267 or at [email protected].
