Embed Notice

@dalias i mean, yes, this would be one, fairly generous, interpretation of the abstract. what i rather see is highly anthropomorphizing language, starting in the title “[…] Reveal Internal Conflict in Frontier Models”. so i very much think these people indeed ascribe these “multi-morbid synthetic psychopatholog[ies]” to the models themselves and not to their makers. and i think its bonkers, because they “argue that these responses go beyond role-play”.