Is Mark Zuckerberg flip flopping on open source AI?

Mark Zuckerberg recently shared a memo outlining Meta's pursuit of AI superintelligence. He suggested that safety concerns might necessitate being more selective about open-sourcing AI models. This stance contrasts with his previous strong advocacy for open-source AI. Zuckerberg acknowledged a potential shift during Meta's earnings call, though he downplayed its significance. He stated that Meta has always selectively open-sourced certain models and will continue to do so. He also cited the increasing size of models, making them impractical for many to use, as a factor in sharing decisions. Furthermore, approaching "real superintelligence" raises significant safety concerns that require serious consideration. Previously, Zuckerberg emphasized open-sourcing as crucial for Meta and developers, arguing it fosters competition and ultimately enhances safety. He believed open access allows governments to monitor and counter bad actors. While Meta will continue open-sourcing some work, Zuckerberg appears to be preparing for a future where their advanced AI might be less accessible.