If you're going to use it, just spend a few hours experimenting and get a feel for what it can and can't do reliably. It's capable of some amazing things, but it also has huge gaps.

I asked it yesterday if it could decode uuencoded text and gave it a sample. It said sure, and decoded it as "Hello world" which wasn't what it said at all. Base64-encoded text, though, it supports and can decode appropriately - but it was equally as confident in its ability to decode both formats.

If you really want to see it freak out a little, try Base64-encoding some directions for it. It'll process them, sort of, but goes very slowly and gets confused between whether it's supposed to be interpreting things or repeating them.