In the competitive landscape of artificial intelligence, Meta Platforms Inc. has been aggressive in its recruitment efforts. However, as OpenAI’s Mark Chen reveals, these strategies have often missed the mark.

High Stakes in AI Recruitment

During a recent episode of the Core Memory podcast, Chen, OpenAI’s chief research officer, discussed the intense competition among AI labs. He noted that while Meta’s efforts to hire top talent have been heavily reported, they haven’t always been successful.

Poaching in the AI Landscape

Chen highlighted that many of Meta’s recruitment attempts focused on his own team members at OpenAI, but they uniformly declined the offers. He remarked, “Before they hired anyone from OpenAI, they went after half of my direct reports and they all declined.”

Zuckerberg’s Unique Recruitment Approach

One notable aspect of Meta’s recruitment drive was the personal touch from CEO Mark Zuckerberg. Chen humorously shared that Zuckerberg even delivered homemade soup to potential recruits. This unconventional method aimed to make candidates feel valued and may reflect the lengths to which Meta would go to secure leading AI experts.

Meta’s Recruitment Budget

With an estimated yearly budget of around $10 billion for AI talent, it’s clear that Meta is serious about its hiring ambitions. However, Chen believes that OpenAI has successfully maintained its core team despite these aggressive attempts.

The Recruitment Competition Heats Up

In the recent past, Meta has made headlines with offers reportedly reaching up to $100 million for talent from rival tech companies. They have successfully hired over 50 researchers and engineers from competitors like OpenAI and Alphabet. Despite these efforts, challenges remain, as demonstrated by the failure to secure notable candidates like Andrew Tulloch from Thinking Machines Lab.

Chen concluded with a light-hearted acknowledgment that he has also adopted the soup delivery tactic in his recruitment, indicating that while strange, personal touches in recruitment could indeed yield results.