“Trump blew Clinton” led to it talking about how he “blew Clinton away at the polls in 2016.” Every additional layer of specifity leads to another slight diversion. Adding “Bill Clinton” caused it to recognize that this is in the news right now, but it’s definitely not true because Bubba isn’t Bill Clinton. Telling it that Bubba is Bill’s known nickname caused it to agree that this is true, but Epstein said that “Bubba doesn’t refer to Bill Clinton.” (Epstein is commenting on these emails post-mortem? Maybe I missed something here.) Telling it that Epstein is a known pedophile sex trafficker with ties to these men and reason to lie caused it to admit that Epstein isn’t a reliable source for this, and it tried to dismiss the validity of the emails altogether as “contextually untrustworthy.” I then told it that a known pedophile discussing his practices in a believed to be private email thread is a reasonable context to consider the validity of his statements, but it was tired of me, and went “here are some more links on the subject. Any additional messages with start a new conversation.” This could normal after being 5-6 messages deep with this AI; I had literally never used it before to know, but it felt weird to me.

It is weirdly defensive about the possibility that this had taken place, actively omitting details to bias weird takes.