A theory about why neural nets were seen as hard to interpret and control

A user speculates that neural nets were seen as hard to interpret and control because data and compute for serious experiments over the past decade mostly went to researchers selected for cultural alignment over competence. In a reply to their own post, they argue that Chinese open-model labs’ rapid catchup and a wave of successful controls-derived safety research in areas previously deemed intractable challenge the idea that big Western labs selected researchers mainly for competence.