First, finishing up day 1, the banquet was, as expected, a fairly nondescript affair involving chicken of some sort, which I happily got to spend with a group of people from EPFL (Lausanne), including Reudiger Urbanke. The big news announced was Shlomo Shamai won the 2011 Shannon Prize, which (as I understand it) is the rough equivalent of the Turing Award for Information Theory.
ISIT, I think, is more like the sort of large conference that Lance Fortnow was trying to push for in his thoughts on re-formatting STOC (as we discussed here and here). I talked a bit with one of the staff who manages the conference (when it's in the US, every other year). This year's attendance was around 750, with over 300 students. He said that was down from previous years, by roughly 10% or so -- 820's was the usual number. His thought was that it was the economy; I opined Austin wasn't the easiest place to get to, especially for people abroad. (Previous years had been in Chicago and Seattle.) At the banquet, they said there was something just under 1000 submissions with about 550 accepted. So they accept over 50% of the papers, but there's still a filter. I'll leave it as a comparison point with STOC -- which has a tougher bar for acceptance, and much less attendance. (And also loses out on several other activities -- see their meeting schedule here. There's a lot of events for students, meetings of various boards, etc. as well as the standard tutorials and plenary speakers.)
The highlight of day 2 for me was going out for Texas barbeque for lunch -- a postdoc I had met at dinner took pity on me when I said I had gone out to a nondescript place by the hotel the day before and been disappointed, and volunteered to take me to an excellent barbeque place she had found earlier in the week. (She was from France, so I accepted her food judgment.) Great food, and we discussed what we liked and didn't like about the conference.
There seemed to be 50+ people in my session when I gave my talk, somewhat surprising given it was the next-to-last session and I'm sure several people had already left the conference. Yashodhan Kanoria gave a great talk after me with some really nice new insights into the deletion channel (paper is on his web page) -- it seems like significant progress in our understanding, as he and his advisor Andrea Montanari seem to have developed insights into what the right "random codebook" looks like, at least for small deletion probabilities. It make me think there's some real chance for significant progress on the deletion channel going forward.