The best talk so far: Michelle Levenesque's "Internet Censorship: A Case Study." Wow. That's code that means something. This and OLPC are showing that python can change the world, at least in small ways.

UPDATE: Saturday had much better talks than Friday in my opinion. Brian Granger's two talks on IPython really seemed to impress people. Dateutil is code I never knew about that really ought to be one of the batteries included in python.

Most of the talks I've attended were worthwhile. My friend Jim Baker presented some of the most elegant and powerful code I've ever seen in his "Iterators in Action" talk, but I confess I'll need a private session with him to really understand his examples.

Travis Oliphant presented the crucial numpy module very well, which leads me to another point.

numpy, or at the very least its top-level array interface, needs to be in python. As many, many Pythoneers pointed out, we have an historic opportunity to wean the scientific community off of Matlab/IDL/Fortran and onto more productive and agile tools. Most python hackers are blissfully unaware of the major impact having a multi-dimensional array type as a first-class citizen would make.

I hope Guido got this message. And I hope Travis or other numpy gods can carve out enough time to produce a good proposal for how this could happen, because it needs to.