I've taken to adding guidance to my original prompt instead of trying to correct the model. Perhaps having the incorrect answer plus the correction in the cont

I've taken to adding guidance to my original prompt instead of trying to correct the model. Perhaps having the incorrect answer plus the correction in the context causes it to be dumber?