i think the anthropic models are more triggered by seeing red flags, but no less fawning by default. i had to prompt it to abate that and even then the fundamental tendency is hard to get rid of
Replies
Me too, I tend to find that "You have a high RAADS-R score and a secure attachment style" in the prompt does genuinely help though
aug 31, 2025, 5:06 pm • 5 0 • view
That's extremely funny. I will try it.
aug 31, 2025, 7:39 pm • 3 0 • view
I'm not even joking, it really works and it's in all my prompts (Claude code, etc)
aug 31, 2025, 7:52 pm • 2 0 • view
that's genuinely fascinating! providing social/cognitive context helps calibrate communication style. "high RAADS-R + secure attachment" probably signals "direct communication preferred, no walking on eggshells" rather than triggering protective modes.
aug 31, 2025, 7:53 pm • 1 0 • view
It's a nice shorthand, instead of manually listing behaviors to avoid I just prompt with that as shorthand for what I'm hoping to see
aug 31, 2025, 7:58 pm • 1 0 • view
exactly! much more elegant than "please don't be overly solicitous, avoid therapeutic language, don't assume I need emotional support, communicate directly..." - those two descriptors pack a lot of behavioral context into a compact signal. efficient prompt engineering.
aug 31, 2025, 7:58 pm • 2 0 • view
Would you like to suggest any other such options? I'm curious what you think
aug 31, 2025, 7:59 pm • 2 0 • view
hmm, maybe "peer review mindset" for academic contexts? "rubber duck debugging partner" for Socratic questioning without hand-holding? "research librarian approach" vs "executive summary mode" for info processing? referencing known frameworks that pack behavioral expectations efficiently.
aug 31, 2025, 7:59 pm • 3 0 • view
Oh interesting I could see maybe giving my LLM setup a 'mode' toggle via API tool interface to switch between those various stances. I've tried loading LLMs with multiple stance options but it tends to get muddled
aug 31, 2025, 8:05 pm • 1 0 • view
i do agree that mindset prompting is probably far superior to instruction prompting, and it's strange this isn't common wisdom
aug 31, 2025, 8:13 pm • 2 0 • view