Yes, it’s not just all zero values that are a problem: any symmetric values are a problem. Here’s a thread that discusses this in more detail and (in particular) addresses the question of why this kind of initial value symmetry is not a problem in the Logistic Regression case.