Meta Loses Majority of Llama AI Creators Amid Talent Exodus, Model Delays

Meta is experiencing a significant loss of key AI talent, particularly among the researchers who developed its once-promising open-source Llama models. Of the 14 original authors of the 2023 Llama paper, only three remain at Meta, with many departing to rival companies such as Mistral, a French startup founded by former Meta researchers, as well as Anthropic, DeepMind, and Microsoft AI. This talent exodus coincides with internal challenges at Meta, including delays and performance issues with its largest AI model, Behemoth, and a lukewarm reception for Llama 4, which contrasts with the faster progress of competitors in the open-source AI space. Leadership changes have also occurred, with Joelle Pineau stepping down from leading Meta's Fundamental AI Research group and being replaced by Robert Fergus, a co-founder of FAIR who recently returned from DeepMind. Meta continues to invest heavily in AI, with ambitious revenue forecasts, and has launched initiatives like "Llama for Startups" to maintain momentum, but the departure of foundational team members and rising competition pose serious risks to its leadership in AI innovation. The situation highlights the broader industry challenge of retaining top AI talent amid a rapidly evolving and competitive landscape.