[Cross posted from LessWrong]

I just gave a guest lecture on potential risks from very advanced AI. I didn't end up having much time to put the slides together, so quality varies greatly between sections - some I pulled from more polished talks I had already given, some I pulled from other sources, and some I wrote up quickly and probably missed half the points I should have included.

I probably should polish it up more before presenting it here, but if I wait to do that I will likely not publish it here at all. So I'm putting up what I have as is, warts and all. Maybe somebody will find it useful.

Feel free to borrow any of it for your own purposes as you see fit. Comments, suggestions, and critiques greatly appreciated!