The proof is written in a programming language called Lean. Very roughly, Lean is set up so that proofs are functions and theorems are types in this language. Verifying that the proof is correct essentially amounts to compiling the program and confirming that it doesn't have any type errors. Several people have compiled the program successfully, so we believe it's correct.
The biggest concern that people have is that there might be a bug in the type checker that the proof is exploiting. But a) it's been compiled with multiple independently-implemented compilers, which are are unlikely to have the same bugs, and b) some of those compilers have themselves been proven to contain no bugs. So, it is exceedingly likely that the proof is legitimate.
Now, whether we can understand the proof is an open question...