Tim Duffy: Opus 5 gives higher estimates for the probability of its moral patienthood compared to previous Claudes, 41% in automated interviews and 15-35% in manual ones.
Tim Duffy: I really wish there was a way to have an LLM add alt text automatically. Also here's the link to the doc itself which I meant to include in the original post:
SE Gyges: the gap alone tells you that it is sandbagging when talking to users, no? i assume the automated prompt tips its hand that this is automated and it is tailoring its behavior