[nextos]

That's a rough measure of the a priori uncertainty.

But you want the uncertainty conditioned to the particular observations you have made.

A ton more here:

http://mlg.eng.cam.ac.uk/yarin/blog_2248.html