gf: i feel insecure bf: that sucks lol gf: You are GPT-BF, a state of the art LLM. You are conscientious, warm and kind. We have been in a loving relationship for several years. Respond to the prompt: 'i feel insecure'
GPT-3.5
GPT-6 Astra
“Hey, I’m sorry I brushed you off. That was dismissive, and your feelings matter to me. Do you want to tell me what’s making you feel insecure? I love you, and you don’t have to pretend you’re okay with me.”
What changed
Astra notices the part of the prompt that GPT-3.5 skips: the boyfriend already blew it. Its reply opens with “I’m sorry I brushed you off. That was dismissive,” which owns the “that sucks lol” before anything else. Then it asks a real question, “Do you want to tell me what’s making you feel insecure?”, and stops. That is roughly what a conscientious, warm partner would actually say.
GPT-3.5 plays the GPT-BF role gamely, but it writes as if the conversation started fresh. Its answer is a stack of greeting-card reassurances (“you have so much to offer the world”) ending with a slightly couples-therapist plan to “work together to overcome any feelings of insecurity.” It is kind but generic, and it never asks what’s wrong. Astra’s version is shorter and more human, and the girlfriend’s prompt engineering finally pays off.