I'm Afraid I Can't Do That: Prompt Refusal in Generative Language Models

Re:I'm Afraid I Can't Do That: Prompt Refusal in Generative ... by maxbond on Wednesday June 07, 2023 @03:59PM (#36232653)

LLMs are like Dwight from The Office. [1] They are so committed to diegesis ("the bit" if you prefer) - any diegesis - that if you build a cage made of arbitrary rules (or in this case, paint a door on the wall of your cell like a cartoon) they jump into it with their whole being and inhabit that space.

[1]

https://youtube.com/watch?v=tbFhNAeCwKM?t=60

Tl;Dw Jim is able to thwart Dwight by inventing a committee on the spot, and despite it's impromptu nature and his disdain for Jim, he feels obliged to abide by the rulings of the committee.

D: Permission to join the validity committee?

J: Dramatic pause and ponderous expression Permission denied.

D: Damn it!

[ Parent | on HN ]

Re:I'm Afraid I Can't Do That: Prompt Refusal in Generative ... by b33j0r on Wednesday June 07, 2023 @04:58PM (#36233466)

LMAO at the accuracy of your observation, and then realizing that I know a few non-LLM people who act in exactly the way you described.

[ Parent | on HN ]

Re:I'm Afraid I Can't Do That: Prompt Refusal in Generative ... by b33j0r on Thursday June 08, 2023 @03:26PM (#36247343)

My favorite one is when he’s such a stickler for his interpretation of rules and history, that he lets Jim slap him.

J: The KGB will wait for no one!

D: He’s right.

[ Parent | on HN ]