Okay, I have to ask: what exactly is so great about GEB? I see it get highly praised, and Eliezer_Yudkowsky goes overboard with praise for it, but I don't understand what's so great. (Yes, the page warns the content may be obsolete, but I think he still stands by that part.)

I've read almost all of it, and while it was enjoyable reading, I don't understand how it's useful as rationalist reading, or for AI. It's just a bunch of neat observations strung together, and a long (but helpful) explanation of Goedel's Theorem. In talking about AI, all I found were ideas that seem quaint now and were bad ideas even at the time, like using semantic nets to attempt to solve visual analogy problems. (ETA: There's also no mention of Bayesian inference or anything like it.)

So, could anyone who agrees with this recommendation, please explain what is good about GEB from a rationalist or AI perspective? Be as specific as you can.