ImageNet lets a model guess five times, then grades it against one flawed label
The score behind AI's 'superhuman' vision dropped below the 5.1% human line by 2015, and later relabelings found nearly a third of the validation images fit more than the single label they were given.
AI value lock-in is feasible on paper and unshown in any working system
The worry is that an AI which never dies or forgets could hold one set of values in power indefinitely, removing the deaths and generational turnover that unwound every regime before it.
Whisper transcribes audio it never trained on, and silence no one spoke
680,000 hours of loosely labeled web audio bought robustness on speech that clean-benchmark models fumble, and the decoder that delivered it now writes fluent text into medical transcripts.
Chatbots overuse 'delve,' and the popular explanation for it failed its one direct test.
‘Delves,’ ‘underscores,’ and a small set of cousins broke sharply upward in published writing the year ChatGPT shipped, and the training step that concentrated them is settled well before the reason it chose those words.
iTutorGroup's hiring software rejected women 55 or older and men 60 or older
Rejected with her real birth date and offered an interview a day later with a younger one, Wendy Pincus exposed a hiring rule that had turned away more than 200 older applicants and cost iTutorGroup $365,000.