Your AI Assistant Wrote the Code. Who Checked the Defaults?

AI assistants often generate code that relies on default library settings, which can introduce subtle issues into machine learning models. The scikit-learn library contains several such defaults that warrant closer inspection, such as RandomForestRegressor using all features at every split, which differs from the standard algorithm. Other overlooked settings include LogisticRegression applying a fixed regularization strength, cross_val_score using non-shuffled data folds, and KMeans performing only a single initialization run. These implicit choices can lead to flawed model evaluation, suboptimal performance, and unexpected behavior when processing new data. Developers must actively review these defaults rather than accepting the code generated by AI tools without scrutiny.