A passing contamination check on GSM8K would not catch a renamed problem
GSM8K's training set went into GPT-4's pretraining before the model scored 92 percent on the test, and a separate study shows that renaming a problem's numbers alone still moves a model's score.
BERT's near-human score on a reasoning test collapsed once the shortcut was closed
Devlin, Chang, Lee and Toutanova's 2018 model set new marks on eleven language tasks by training on masked words instead of the next one. One of its most human-looking scores, though, traced to a handful of words a benchmark let it exploit, not the argument the task was built to test.
No AI has to go rogue for humans to lose control
Its authors admit they can't yet show how the loop closes, and the starkest real cases so far, two automated Dutch systems, ended with a court striking one down and a cabinet falling over the other.
A model's weights freeze the moment its training run ends
Everything a chatbot seems to learn afterward, from a correction to a live search result, arrives through the prompt: its weights were never touched again once training ended.
IBM trained Watson for Oncology on hypothetical patients, then sold it worldwide
Its own internal reviewers called the results unsafe and incorrect, and the same slide decks weighed the business risk of finally testing them.