- Witten's Langlands talk at Harvard
- Witten's previous review of the topic, with some relevant links to CFT papers by Strominger et al. and Vafa et al.
- Kapustin's comments about S-duality and exceptional groups
A semi-relevant discussion about related topics occurs at Not Even Wrong.
In the article, one learns that most of the miraculous results in mathematics that go under the name "geometric Langlands program" follow from a small portion of a decoupling limit of a topologically twisted background of string theory. ;-)
In the fast comments under this text, X.Y. argues that the non-trivial Langlands program is the arithmetic one, not the geometric one.