Prof. Wilson's talk, The Bits of Evidence, got the most applause from the audience. It was a great idea to allocate one slot to an "academic" speaker so that we all could step back from products, languages, and frameworks and look at the big picture. Greg Wilson showed numerous examples of how the software field is not up to the same standards of scientific proof as evidence-based medicine and how things are changing.
The Python talk (by Jordan Baker) was also very engaging. The speaker could have easily spent one hour going over the syntax of various language statements, but instead, fortunately, he chose one problem that is not trivial, but can be solved effectively using the unique powers of Python. I am referring, of course, to the spelling-correction problem ("did you mean...?") and Norvig's algorithm. The speaker then built the algorithm from ground up, line-by-line. That was a great introduction to Python.
I am sure Reginald Braithwaite's talk on metalinguistic programming got some people in the audience to rethink their thinking about thinking about thinking about programming. The entire presentation was written in Ruby syntax, which was very very cool.
"Frameworks talks" (on ASP.NET MVC and jQuery) were not as exciting, but contained a lot of useful information and the speakers definitely knew their topics.
Joel's keynote was quite entertaining: many examples of how to simplify a software product by removing decisions that annoy users, the acknowledgement that complex, feature-rich products make more money, the dilemma between simplicity and the need to make money, and the attempt to solve it by picking "good" features over "bad" ones.
And finally, Joel also talked about Fog Creek and its products during the conference. To me that sounded more like information (StackOverflow Jobs, FogBugz's code review features, etc. etc.) and less like a sales pitch.