Why we still has to handle the `Failed to initialize NVML: Driver/library version mismatch` in 2024?

I just followed the CUDA 12.1.X installation guide and successfully set up my research workstation to accept GPU workloads.

However, we often face the Driver/library version mismatch. Is Nvidia unable to make a stable enough driver for Linux? Or will I have to reinstall the drive every time this happens?