I can't readily think of something I used to strongly believe but later strongly disbelieved. Taking what I can, here's my compressed pellet of wisdom:

"Graduate students often tell Michael Wilson they think they know how to create smarter-than-human artificial intelligence that won't fail to care about us. They are probably always wrong, as Eliezer Yudkowsky could tell you. Don't bite off more than you can chew; you're prone to this. By the way, the interlibrary loan system is awesome."

I hope all of those are statements that I would understand and agree with well enough. If I remember correctly, my reaction to the whole Friendly AI idea was, "Oh yeah. Heh."