Over the past few weeks, I’ve been developing and testing a prompt interaction method I’m calling “Your Take / My Take” — a structured calibration exercise designed to shift ChatGPT from generalized cultural fluency to individual resonance.
Most recommender systems — from music to film to art — surface what’s popular, safe, or known. This is fine for passive consumption, but it fails when the goal is discovery. LLMs are not recommender systems. They are pattern-matching scaffolds, capable of identifying tensions, contradictions, and preferences based on behavioral imprint, not surface data.
Why this matters:
In testing this with GPT-4, I’ve found that with just a handful of conversational turns — where I offer opinions on a few cultural artifacts (films, TV shows, albums) — the model quickly begins mirroring back not a profile, but a behavioral interpretation of how I engage with media, aesthetics, and ideas.
Real-world impact:
- I was recommended the TV show Patriot based on emergent patterns — not because it’s trending, but because it aligned deeply with the contradictions in my media tastes. I never would have discovered it through traditional recommendation channels. It’s now a personal favorite.
- I used this same structure to explore the Cincinnati Art Museum, and was pointed toward 15 specific works that resonated with my underlying ethos — including pieces that both affirmed and challenged my aesthetic and emotional leanings.
This isn’t about building a psychometric profile. It’s about activating discovery — using an LLM as a mirror, not a guide. Using the LLM as a system of resonance not intelligence or reasoning.
The prompt structure is simple:
I give a piece of media. You (LLM) give your take based on cultural context. Then you give your interpretation of my take, based on what you know of me. I respond. We repeat. Calibration sharpens. Discovery follows.
Why this works:
LLMs are trained on collective language patterns — but when we interact through pattern-based calibration, we flip the system. We move from algorithmic slop to personal signal. From parrot to companion.
This prompt framework can be used with media, art, music, even public institutions (like libraries and museums) as a portable scaffolding for guided, intentional exploration. It requires no code, no app, no extra infrastructure — just the right kind of conversation.
This is part of ongoing thought around not being anchored to our devices and allowing them to be instruments of distractions but rather engines for discovery. How we can move beyond our hardware and using our tools for accomplishing goals, fostering ideas and living better and more meaningful lives.
Would love to hear if others are experimenting with frameworks like this — or if any LLM teams are building reflection-calibration loops into their designs.