I'm way too late for this to be meaningful, but here's what I think! tl;dr -- gravity is the problem, we should focus on experiments and observations for a while, bring in some better mathematics, and continue the long range program of developing quantum theories that include or even produce GR.
I'm just a physics enthusiast. When I became interested in physics, I was initially a sort of partisan in the "realist" camp -- pro Einstein, anti Bohr; liked Verlinde's entropic gravity, distrusted the graviton -- but have come full circle to the opposite view. GR has massive explanatory and predictive power, and an extremely satisfying aesthetic quality, but obviously breaks down behind the curtain of a black hole's event horizon, where we cannot make observations. I say obviously because it predicts a singularity, which is just another way of saying it makes no prediction at all. On the other hand, many of my complaints about QM I now look at as unanswered questions, opportunities for inquiry. QM is based 100% on experimental observations. The theory came together in a rather ad hoc fashion at the beginning of the last century, but as it was more carefully studied theoretically and experimentally, also proved to be highly predictive, even more that GR. Yes there are big ugly, outstanding questions -- measurement collapse, the transition from microscopic quantum behavior to macroscopic classical behavior, the intractability of all but the simplest calculations -- but those are huge areas of knowledge that the future will gradually (or suddenly) fill in, as our understanding moves forward.
So, gravity, not QM, is the problem. We should start with the axioms of quantum mechanics, and look for ways to observe where QM and GR can be measured at the same time. LIGO offers opportunities like this, as the sensitivity of the instrument is well within the quantum regime. Continued study of QCD could make a testable prediction for what exists inside of a black hole. Or continue to study the very fine transitions between energy levels in the nucleus. Mathematically, maybe the Langlands program, with its rather weird, Fourier-like sums of L functions will allow us to model non-linear behavior.
And yes, string theory sort of "jumped the shark" at some point. I'm sure Ed Witten regrets saying that other pursuits were a waste of time. But the thing is, the string theory program is centered on QM, and has shown that QM can naturally produce GR, given certain unrealistic assumptions. That's mathematical progress.