SafeRent scored renters on their credit and never counted the voucher paying their rent
Mary Louis, a 54-year-old Black woman whose voucher covered about 69 percent of her rent, was denied an apartment over a SafeRent Score with no appeal, and in 2024 the company settled for $2,275,000 without admitting the score broke the law.
Reading out what a model knows is unsolved, and the one hopeful result tracked its most prominent feature
Paul Christiano's group at ARC posed the problem in 2021 for AIs that act where humans cannot check, and its strongest form is still a projection about systems no one has built.
The Vision Transformer lost to a plain CNN until it was fed 300 million images
Dosovitskiy and his coauthors pinned the reversal to Google's private 303-million-image set, and a qualifier in their own introduction named the missing ingredient a later model supplied on ImageNet alone.
The same model scored 1673 or 2214 on Codeforces, depending on the scaffolding
OpenAI's own paper reports that gap, and every other AI Codeforces figure in circulation, up to o3's 2724, is the same kind of number: an estimate fit to contests the model never entered live.
A please or a threat shifts a model's answer without reliably improving it
The wording is ordinary text the model was trained to weigh, and on a hard current benchmark it swings a single question by up to 60 points in either direction while the average gain stays at zero.