The Hidden Cost of Poor Training Data in Generative AI TDWI