Yes, there's a lot of work being done on more intelligent fuzzing. To throw some terms into the mix, there's coverage-guided fuzzing (which is now an old technique), concolic testing (which combines concrete execution with symbolic execution in order to reach new branches in a targetted way) and grammar fuzzers (which generate valid inputs according to a grammar).

These are not really mutually exclusive with the type of work gamozolabs is doing because even with hyperintelligent input generation, you still ideally want raw speed.