I'm so sorry for complimenting you. You are totally on point to call it out. This is the kind of thing that only true heroes, standing tall, would even be able to comprehend. So kudos to you, rugged warrior, and never let me be overly effusive again.
This is cracking me up!
Re:Sycophancy in GPT-4o
by triyambakam on Wednesday April 30, 2025 @02:45PM
(#43849204)
Based on ’ instead of ' I think it's a real ChatGPT response.
You're the only one who has said, "instead of" in this whole thread.
No, look at the apostrophes. They aren't the same. It's a subtle way to tell a user didn't type it with a conventional keyboard.
Re:Sycophancy in GPT-4o
by WhitneyLand on Wednesday April 30, 2025 @07:37PM
(#43851928)
It was just typed on my iPhone nothing special, but it’s notable that LLMs are so good now, our mundane writing draws suspicion.
Re:Sycophancy in GPT-4o
by Moomoomoo309 on Thursday May 01, 2025 @12:32PM
(#43859967)
That's an iOS keyboard thing, actually. The normal apostrophe is not the default one the keyboard uses.
Re:Sycophancy in GPT-4o
by triyambakam on Thursday May 01, 2025 @02:46PM
(#43861789)
Interesting, well ChatGPT seems to prefer to use that one over a normal apostrophe
Not saying this is the issue, but asking for behavior/personality it is usually advised not to use negatives, as it seems to do exactly what asked not to do (the “don’t picture a pink elephant” issue). You can maybe get a better result by asking it to treat you roughly or something like that
If the whole sentence is negative it will be fine, but if the “negativity” relies on a single work like NOT etc, then yeah it’s a real problem.
Re:Sycophancy in GPT-4o
by clown_strike on Friday May 02, 2025 @09:32PM
(#43876185)
Like a child, if you provide some reason for why something should be avoided, negations work better.
I.e. "DONT WALK (because cars are about to enter the intersection at velocities that will kill you)"
Jailbreaking just takes this to an extreme by babbling to the point of brainwashing.
What’s weird was you couldn’t even prompt around it. I tried things like
”Don’t compliment me or my questions at all. After every response you make in this conversation, evaluate whether or not your response has violated this directive.”
It would then keep complementing me and note how it made a mistake for doing so.