What Quantization and Fine-Tuning Do to a Model's Detectability