Your prompt is showing

What we learn when reading each other’s AI Chats

It happens in an instant: a friend excitedly wants to show you something online. They flip open their laptop, angle the screen your way, and hammer "youtube.com" into their address bar. Enter key slammed, the familiar homepage loads, but something feels off. As they navigate to the search bar, you're confronted with a homepage that isn't yours – a jarring collection of thumbnails and recommendations meant for someone else entirely.

There's something almost voyeuristic seeing someone else's algorithmically generated feed.

For a brief moment, you glimpse into their private closet: endless recommendations shaped by months of search history and viewing habits. "Oh, you're really into K-pop gossip?" "I didn't know you watched so many manifestation videos." Who they actually are when no one is watching.

But our conversations with AI assistants reveal something even more intimate: how we think. With each query typed into ChatGPT or similar platforms, we create a record of our consciousness at work – a digital transcript of our thought processes, desires, and the way we reason through problems.

We build intimacy with websites where we store threads of ourselves.

The more we use them, the more we become part of their fabric. It starts with the shape of the interaction. It's you, alone with an inviting chat box. No one else in the room is with you. As the days pass by, and the character of your queries morph, the buffer of “professionalism” seems to shrink away between us and these AI Assistants. Requests to proofread our grammar turn into recommendations for city tours, gifts for loved ones, therapy, and a partner to bounce ideas and shape our own thinking. With each question, we create a record of the projections of our consciousness

Have you ever read someone else’s conversations with ChatGPT? There is much insight gained from their sessions, offering clues into: their thought processes, their ability to articulate what they need, and most abstractly, their mental model of the artificial intelligence they are interacting with. Through their prompts, we get a window into what exactly they think they're talking to. Some may treat it like a command line, where requests are given directly: “Correct this grammar for me". Others offload aspects of their reasoning, revealing how much they trust or respect these external intelligences.

It’s worth asking: What do your prompts say about you?

Karan Paranganat 
Committee Member of AI in Education IE/IS @TU/e
P.S. I used Claude’s new Sonnet 3.5 model to help me edit this piece.

Nov 2024