Machines cleared SQuAD's human exact-match score without learning to read
Microsoft and Alibaba passed the exact-match line on Stanford's reading benchmark in January 2018 while still trailing humans on F1, months after a single added sentence had halved the scores of the field's best reading systems.
Post-training makes a model less able to tell when it's wrong
A base model's own probabilities track how often it is right, and the confidence a chatbot types out was never a reading of them.
Competition could select for the AIs no one wants, and nothing is selecting them yet
Dan Hendrycks argued in 2023 that competition, not any lab, would breed power-seeking AI, and even his sharpest peer-reviewed critic grants the danger turns real the moment AIs compete with no human deciding which survive.
Michigan spent nearly two years accusing jobless workers of unemployment fraud with no human examiner
MiDAS flagged fraud on automated data mismatches and imposed a penalty of four times the benefits, and the state's own 2017 review later reversed 85 percent of the findings it had made by computer alone.
Adam's convergence proof was wrong, and the field kept using Adam
The 2018 counterexample that broke the proof drives Adam to the worst point of a one-dimensional convex problem, at hyperparameters no practitioner ever chooses.