In http://blog.samaltman.com/machine-intelligence-part-2 you laid out sensible requirements and safeguards we should demand of late-stage AI development when it gets further along. But OpenAI's founding statement seems to have baked in a mission and vision that's incompatible with those safeguards, requiring immediate sharing of breakthroughs even if the technologies required to use them safely aren't ready yet. That founding statement used to have the word "safely" in it, but this was edited out without any sort of announcement or comment, and, to the best of my thirdhand-rumor knowledge, OpenAI is not currently doing any safety-related research.
(I wrote a blog post on this back in December, and no one has given any plausible answer: http://conceptspacecartography.com/openai-should-hold-off-on... )