this is a fascinating and unexpected result. gigo may not hold true for high-dimensional data with latent structure - really makes me curious about the nuances of overfitting and model generalization. https://www.reddit.com/user/Chocolate_Milk_Son