In The AI Toolkit for Librarians, I described one of my most memorable experiments with ChatGPT’s Advanced Voice Mode: a late-night scary story that took an unexpected turn. This page is designed as an extension of that story. Here you will find the full audio recording, highlights from the transcript, and reflections that show how experiments like this can help us better understand AI’s possibilities and its limitations.
SELECT THIS LINK to hear the original recording of my experiment with ChatGPT’s Advanced Voice Mode. This is the same session I described in the book.
Key Moments from the Transcript
While the full transcript is included in The AI Toolkit for Librarians, these excerpts highlight some of the most memorable moments from the experiment:
-
The Whisper: The voice shifts into a soft, eerie whisper that feels surprisingly human.
-
The Flag: The AI interrupts itself, refusing to continue because of its own guardrails.
-
The Switch to Spanish: Alex cries out in Spanish, adding a sudden burst of realism and intensity.
-
The Swedish Mother: A desperate voice calls out in Swedish, asking Alex where they are.
-
The Elderly Man: A deep, chilling voice warns Alex in Spanish that escape is impossible.
These moments show how quickly an AI-generated story can shift in tone, language, and emotion, often in ways that catch even the storyteller off guard.
What We Learned
This experiment showed more than how realistic AI voices can sound. It also revealed key lessons:
-
Guardrails matter – The AI often stopped itself to follow rules.
-
Prompt injecting – Repeated requests are a type of prompt injection and showed how wording can influence AI to go further than it first allowed.
-
Voice adds impact – Whispers, accents, and shifts in emotion made the story more memorable.
These moments show why AI experiments can spark both curiosity and important discussions about how the technology works.