In the video you are rotating the arm at an extreme angle. Anatomically speaking, this wouldn’t be possible in a real arm. I know that doesn’t matter when we’re just trying to animate a 3D mesh. Or if the rig needs to do specific non-anatomical things, like robot rigs.
But if you’re raising the arm up that high, you should likely be rotating the clavicle up as well. And then parent-relatively, the arm won’t be passing that 180º flip zone.
If rotating from the clavicle doesn’t look right, you could also try giving the arm an extra parent at the same pivot, and rotate it up in the same way as you would the clavicle. This would give your driver an extra bonus gimbal. It would take a bit of extra manual animation, but it’s better than counter-rotating an unstable gimbal issue.
The only other thing I can suggest is tracing along the DAG and finding the drivers. Like arm_L0_div0_loc
and attempting to change the rotation order. But if you do that, it will flip during other rotations. Euler rotations and upvectors can only do so much.