Very cool direction…thanks for dropping it here
1 Like
Just published my second persona: Electric-Shy
The opposite of Cheeky Razor, quiet, hesitant, analytical. Fragmented reasoning instead of sharp quips.
https://github.com/LarisaHaster/Electric-Shy
45 tests, 100% pass rate (better than Cheeky Razor). Works on Claude and OpenAI.
Enjoy!
1 Like
Hi everyone!
I’m experimenting with a sharp-tongued AI persona that should keep a sarcastic / blunt tone.
What’s the best way to structure the system prompt so the model stays in character across different GPT versions without sliding into polite neutrality?
Are there known tricks to anchor personality while keeping output accurate and flexible?
It’s my understanding that certain models are simply going to be polite, as the snarky-off-the-cuff stuff basically cuts across the grain too much.
While there might be workarounds that I haven’t noticed or been made aware of … it’s my understanding that such is handled with models that don’t fight the ruleset in the first place.
Which GPT versions are best suited for maintaining a sharp-tongued / sarcastic persona without sliding into polite neutrality?
Are there known tips or tricks to anchor this style reliably?
my thing was to use known actors who played good roles as such to anchor to, but the deeper levels require constructing personality skill files.
All of this was covered in the thread above, best to simply scan through it and find what you need, probably.
For deep personality consistency, what kind of “skill files” or structures work best to maintain a blunt, witty style without losing accuracy?
I’ve indicated that i’m not going to recite the whole thread for you.
The information is here as far as people on this forum have explored it already, but you have to read the thread yourself.
1 Like