Attractors in end-to-end neural diarization may not need to encode speaker characteristic information.