Should powerful foundational AI models be kept under lock and key, or shared openly with the world?

In this episode of Approximately Correct, we sit down with Joelle Pineau, a professor at McGill University, former head of AI research at Meta and current chief AI officer at Cohere.. She led the development and release of the early versions of Meta’s Llama models, a series of open-weight models that challenged the closed-door approach of other AI teams. 

Pineau argues that openness is the fastest path to better, safer AI, and that diversity in foundational models is essential to developing a limited ‘algorithm monoculture’.