Is everyone aware that even with Memory and Personalization turned off, ChatGPT’s system prompt started to include your name *and* email on every chat?
Seems like a bad idea that you can’t turn off?? It biases all responses and creates more attack vectors for prompt hijackers.
No Digg Deeper questions have been answered for this story yet.
Not sure when it started being the case. I first noticed that maybe 2-3 months ago. Claude doesn’t seem to do this. I might be missing something?
Anyway, by “biases all responses”, I mean that the model is forced to attend to who you are when answering even the most generic of questions. Not always a desirable property when you want its most unvarnished answer let’s say.
Is everyone aware that even with Memory and Personalization turned off, ChatGPT’s system prompt started to include your name *and* email on every chat?
Seems like a bad idea that you can’t turn off?? It biases all responses and creates more attack vectors for prompt hijackers.
@lateinteraction why include name and email in the system prompt if users can turn personalization off?
seems like those two things should align.