This is great news! I think distributed access, control, and contribution to the best AI's will help create 'safe' AI's much faster than any AI created in secret. One thing this does not address, and is something that Jerry Kaplan has an excellent suggestion his recent book "Humans need not apply", is the distributed ownership of AI where tax incentives to public companies that have larger numbers of shareholders, encourages wider distribution of the massive gains AI will bring to these companies.

I really hope that the training data, as well as code and research, will be opened up as well, since the public could really benefit from the self-driving car training data Tesla may contribute[1]. By opening up the development of this extremely important application to public contribution and the quality benefits that it brings, we could get safer, quicker realization of this amazingly transformative tech. As of now the best dataset for self-driving cars, KITTI, is extremely small and dated. [plug]I am working on a project to train self-driving car vision via GTAV to help workaround this (please contact me if you're interested), but obviously real-world data will be better in so many ways.

[1] https://medium.com/backchannel/how-elon-musk-and-y-combinato...