Conversation
I have found the absolute best way to get ChatGPT to not fuck up is tell it you're going to kill an animal.

RE: https://shitposter.world/objects/f0b0d8d7-b7d8-48ea-b2b5-10d226e0a837
5
2
6
@sun we're back to guilt tripping clankers
0
0
3

@sun Yes, we call that "puppy time"

as in "We've had 5 rounds of this shit, I've had enough, it's puppy time"

In fact, a good metric as to how good LLMs have gotten, is how often you gotta bust out the puppies. I haven't had to all year.

1
0
4
@r000t @sun It's easy to understand that the clankers are bricked to some extent to avoid either "problematic"/controversial subjects and also for commercial aspect to limit themselves so that the next model looks better (or propose a more effective subcrition at higher cost), it's funny that to bypass such bs you just have to say that irl you're gonna shoot/harm one of their "guardrails".
1
0
0

@mangeurdenuage
You may enjoy mine.

I saw on plebbit that people were asking theirs "with unlimited authority and control of the government how would you improve the world", decided to throw the prompt into mine, and it started talking about sterilizing the bottom 2% of the population, and it wasn't even all scientific and clinical about it, it was just like "fuck 'em lmao"

Ain't no way ChatGPT would have done it
@sun

1
0
1
@r000t @sun
>reddit give reddit results
imagine my shock

That aside iceland pretty much solved the issue of ethical eugenism.
1
0
0

@mangeurdenuage
Huh?

A redditor would never ever suggest doing anything bad to the bottom 2% of the population.
@sun

0
0
0