The way we train AIs makes them more likely to spout bull

The tendency for AIs to give misleading answers may be in part down to certain training techniques, which encourage models to prioritise perceived helpfulness over accuracy

​The tendency for AIs to give misleading answers may be in part down to certain training techniques, which encourage models to prioritise perceived helpfulness over accuracy The tendency for AIs to give misleading answers may be in part down to certain training techniques, which encourage models to prioritise perceived helpfulness over accuracy