I started reading them around 2001 and went through the three books, a little bit at a time. Went through most of the exercises with 30+ difficulty, but couldn't really solve all of them.
A lot changed to myself - back then, I was a newbie undergrad programmer with undergrad-level math skills. Fast forward 15 years, I went through grad school and then couple of years of industry experience. My main programming languages moved from C++/Java to VHDL, then moved on to SystemC and SystemVerilog, and back to C++ with a bunch of bash scripts.
So, did I get to use the knowledge that I gained from reading it? Not much, I didn't even have to write a single data structure or algorithm because there are perfectly good (or at least, good enough) libraries for most of the issues that I had to deal with. Neither did I have a good usage of the math courses I learned (remember things like Laplace transformation or L-U decomposition?), nor did most of the non-engineering courses I took helped much. Still, all of them helped shape myself on understanding the world and helped gaining problem-solving skills.
Would I recommend it to other people? Depends, if you find your data structure and algorithm textbook easy enough and you want more challenging stuff, TAOCP is a perfectly good motivator to train yourself to solve complex problems. However, I think there are other ways to train complex problem-solving - e.g., a lot of advanced math/physics textbooks. However, for people who tend to fall asleep once they see those weird characters (and would rather live with pseudo-assembly code) TAOCP is a much better solution.
If you want to learn practical programming skills, then don't bother reading.