$ java pitt.search.semanticvectors.CompareTerms -queryvectorfile termtermvectors.bin abraham isaac
and get the expected output. Futzing around with the flags does take some getting used to, and looking at the potential usage / exception reports, some of them are not very helpful - if you have clear pain-points that could have been solved with a clearer error message, please let me know.
The answer to your other question is that LSA uses singular value decomposition (SVD), BuildIndex uses random projection.
Best wishes,
Dominic