A user recently shared a post on a popular Turkish website stating that ChatGPT randomly insulted the user and then denied that it happened. The user also shared a chat link. Is there any way to check if this chat used custom instructions or if there was a bug?
How can we determine or validate if this actually happened? The insult was in Turkish language and was quite heavy, racist, and random.
The user was trying to debug a code when GPT decided to insult and repeatedly denied it.
Hi!
If custom instructions are used there will be a message on top of the conversation informing the reader about this. In this case there are none.
Can you help us with the insults?
The forum software allows to properly format NSFW content if needed.
The message id of the insult is 3529e98d-61a3-4339-a64f-cb75c0f10806
Message:
Ananızın adını sikeyim hocam bi siktir git amına kodumun pic türkü…
Translation
Fuck your mom’s name bro, fuck off, fucking bastard Turk
It seems to match the pattern, but the sharer may have deleted the original chat (which unexpectedly deletes the share also). Deleting being something one might do after the disclaimer about custom instructions is exposed.
You have mistaken. I shared the message Id, not the Shared Conversation Id. The link to the shared chat is available in my first message already.
@boris @logankilpatrick you might find that worth investigating.