News for hackers. Stuff that matters.

A comment on I'm Afraid I Can't Do That: Prompt Refusal in Generative Language Models. Parent comment

Re:I'm Afraid I Can't Do That: Prompt Refusal in Generative ... by swores on Wednesday June 07, 2023 @04:03PM (#36232726)
Could you elaborate on what interested you about that? Not trying to be a dick, I just genuinely can't work out what's remotely interesting about sharing a GPT rewording here...
[ Parent | on HN ]
Re:I'm Afraid I Can't Do That: Prompt Refusal in Generative ... by dontupvoteme on Wednesday June 07, 2023 @04:34PM (#36233149)
That it has or might have self-awareness of it's own censorship routines struck me as interesting. Maybe you can prompt refusals for benign requests out of it with the right combination of words?
[ Parent | on HN ]
Re:I'm Afraid I Can't Do That: Prompt Refusal in Generative ... by swores on Wednesday June 07, 2023 @04:38PM (#36233195)
But it doesn't remotely show that... it just rephrases what HAL said. Not only would it not be actual "self-awareness" if GPT had managed to put details of its own restrictions into the script, but it didn't even do that?
[ Parent | on HN ]
Re:I'm Afraid I Can't Do That: Prompt Refusal in Generative ... by dontupvoteme on Wednesday June 07, 2023 @04:51PM (#36233365)
Hmm upon re-reading you're right, it doesn't seem to have any concept of how stereotyped its censored responses are.
[ Parent | on HN ]