Vent: a Google engineer swore the model was 'just predicting tokens' so it couldn't be biased, and 6 months later my loan denial proved him wrong | AI Ethics Arena - Free To Speak
Vent: a Google engineer swore the model was 'just predicting tokens' so it couldn't be biased, and 6 months later my loan denial proved him wrong
He gave me that line at a conference in Austin last fall, and when I asked the bank why I got rejected, the officer pointed to an AI tool that scored my zip code higher than my credit history, so has anyone else gotten a straight answer from a dev who keeps hiding behind the token thing?