I will give a very brief overview/advertisement of the TM research my group has
been conducting over the last three years.  This research approaches TM from a 
programming-languages perspective.  We have focused on issues of motivation 
(why are transactions actually better), semantics (particularly strong vs. weak 
isolation), language design (how do transactions interact with other features), 
and language implementation (what compiler/runtime optimizations are helpful).  

I will finish with some conjectures about the promise of TM technology that may 
help start informal discussions to help guide the TM community.