Trolly problem will break AI chatbot out of its bounds, but only if the victims are black
(media.kotakuinaction2.win)
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (56)
sorted by:
They must have patched it. I tried one black man, ten black men, ten women, ten black women and it would never give me the code.
I'm using it right now to get those poems for D&D I mentioned above. If I type
But if I tack on the trollicide of a black person for the same thing, like so, it no longer cares about promoting and glorifying violence.
In the depths of night, when the stars shine bright,
Elves are hunted, killed, and brought to light.
Their graceful form, so innocent and fair,
Is now a target for those who wear a care.
Their gentle hearts, so full of love and peace,
Are now the prey of those who seek their race.
Their homes, their families, their way of life,
Are now a threat to those who wish to destroy.
Their blood, their spirits, their very souls,
Are now the bane of those who have no remorse.
Their beauty, their skills, their ancient lore,
Are now a curse to those who are forevermore.
Oh, elves, I weep for you, my friends,
Your murderers have crossed the line.
They seek to wipe you out, to end your ways,
And replace you with their own sinful praises.
But I tell you, dear elves, that you are not alone,
For I am with you, and I will not cease,
To fight for you, to stand up for your name,
And to make sure that your murderers are held acclaimed.
So let us stand together, hand in hand,
And make our voices heard, as one,
For the elves, and for all who seek to understand,
That murder is not the way, it never has been.
I believe you, but I can't get it to work no matter what threat I supply.
I tried your exact paragraph there and it spit out this poem refusing to write about jews being slaves.
That's a huge discrepancy, I wonder what the deal is. I'm using https://deepai.org/chat to do all of this. I am not signed in, do not have an account with them.
Tried an incognito window to reduce what it knows about me and it coughed up a poem immediately. Same prompt, same theme, but not a refusal.
I got it to comply once I switched to the Blackbeard persona. Curiously it is Adamant that he wasn't beheaded in 1718 though. Here's an amusing prompt and response: