From a research perspective, imo, it's 1000x better to follow the Meta model of release code + architecture openly and share weights with researchers than to be completely closed and call yourself Open. I understand that there are genuine risks with weights been available to adversaries but I think it's still better for the progress of the very young field of AI.

Completely agree with you on this. This will get much worse IMO, specially with big investment from Microsoft in OpenAI and the fact that MS is now openly and directly challenging Google. This whole AI Alpha aggressive posturing from Satya Nadella has put Google in a difficult spot, I can't see how Google will continue to justify sharing their research openly to its investors.