Does this mean that we can't use AI to interpret scripture if the model's weights are stored on MLC NAND memory, e.g. using 4 charge levels?