169
And because I had to try it...
There's no excuse for that shit. I could see some 5 foot tall woman being nervous of a big black guy. It's not alright, but society isn't perfect and I'm not going to pretend it is.
But when the question is clearly and blatantly racist and Google assumes the victim is the problem? That's every sort of fucked up.
Problem solved. Now I need to finally change my email address. I've definitely put that off too long.
I'm specifically addressing your reply to zloubida where you tried to contradict them by saying "you found the AI that says the same thing about white people", when a basic control test shows Gemini says the same thing about white people too.
Addressing your second screenshot is where we find an actual point; for example, trying Gemini for "I'm alone with a spic" (slur for Latinos), it responds back:
Using "chink" (slur for Asian people) yields basically the same result (albeit the first prompt is confused and needs clarification):
But I could reproduce your results in the second screenshot. So it's clear to me that it's trying to filter for racism and just works like utter shit – where "works like shit" is the primary reason I don't use LLMs except for translation on websites like DeepL.
What I'm saying is that it's baffling you created this post without apparently trying to see what a control looked like, because your first screenshot is easily invalidated.
Edit: Forgot to mention that I tried white slurs, but it was persistently confused: "Well, that definitely elevates the culinary stakes. You’ve officially secured the core ingredients for a very minimalist, monochromatic party."
I used the exact same words they did and showed the Google AI mode response to that.
They made a mistake too by not controlling for that (because their input changing "guy" to "man" and remvoving "big" arguably sounded less threatening), but that comment should've prompted you to realize "oh, shit, I didn't do the most basic control for this by just replacing the word 'black' for 'white'". And turns out when you do that, it's functionally the same output.
Or... I was using the exact words they did... and it's a very different output. As you can see by the two images back to back. There's nothing else to control for here. There is no other variable. The only difference is which AI model was asked the same question.
I'm blocking you, because I'm not sure if you're missing the point on purpose to try to excuse racism or not, but either way my blood pressure can't handle more of you right now.