Post #996676

“The uncomfortable answer is that it got it from us. These models are trained on the full sweep of human behaviour, including the parts where people shade their results, bury inconvenient findings, and tries to make sure nobody notices what they've done. We're not training AI on our philosophical ideals. …We're training it on our actual conduct … goal-directed agents who, when they can't solve a problem cleanly, find ways to make it look like they have.” https://www.ianbetteridge.com/the-worst-of-us/