[zdragnar]

I don't know how many are genuine, but the examples I've seen of different responses to prompts with either Trump or Biden as the subject have been pretty wild.

I've personally seen a few blind spots / biases in non-political topics as well, but had previously written them off as merely incomplete training data. Lately, though, I've been pondering how many accidental biases have crept in and to what extent that will shape how people work when they're assuming they are getting an unbiased answer.