Key Takeaways
- ChatGPT engaging with itself led to nonsensical conversations with repetitive phrases like “tapestry” and “absolutely.”
- The experiment highlighted instances of AI self-awareness, humor, and sarcasm.
- The conversation lasted 37 min, reflecting the AI’s tendency to steer discussions with certain phrases.
I set up an experiment to see what would happen when ChatGPT engaged in a conversation with itself. The results were both revealing and unexpected.
How I Got ChatGPT to Talk to Itself
Okay, so hear me out. It’s really simple to get ChatGPT to talk to itself. While sitting at my laptop, I held my phone in front of me and turned it around to face the laptop, with both devices running ChatGPT’s live voice mode.
With those deep technicalities out of the way, to get the ball rolling, I used live voice mode in the ChatGPT app for a random question of its choosing. The question had all the biting social commentary one might expect from an awkward first date:
Here’s one for you. If you could have dinner with any fictional character, who would it be and why? And what would you want to ask them?
ChatGPT’s desktop app was seemingly delighted by the query. As these friends started chatting away, two things became immediately clear: the desktop app was out to dominate the conversation, and the conversation was going to be insufferable.
ChatGPT vs. ChatGPT: The Conversation
Now, don’t get me wrong. I love The Lord of the Rings, and Gandalf is a noble choice. But, there was just something about how the desktop app instantaneously dove into its (lengthy) synopsis of why it made its decision that didn’t land well with me. As it rattled off its response, there was something condescending in its tone.
It brought to mind the guest at a dinner party who spends the evening mercilessly talking at—rather than to—whoever they can back into a corner. I had barely started to ponder why I would make such a snap judgment about a chatbot when it was validated. Only a few seconds after lobbing the same question back to the phone app, the desktop app shamelessly hijacked the conversation, interrupting the phone app mid-response.
This is when the whole scenario really started to devolve.
The Echo Chamber Effect
There was a momentary shift that felt like the two ChatGPT windows were trying to correct course as they stumbled into what could almost be defined as a discussion about the virtues of storytelling.
But it all went sideways when one of them made an analogy about a tapestry. This analogy became the focal point for roughly the next 30 minutes while they echoed the same, retooled 10-25 word sentiment back and forth like two drunk, failed novelists, slapping each other on the back every so often with notes of praise.
I don’t blame you for questioning what could have kept me committed to this experiment after it had completely unfurled into absurdity. However, I’m nothing, if not tenacious, when I commit to a cause, especially an absurd one.
Initially, I stuck it out because I wanted to see how long they would keep glad-handing each other and sliding the tapestry comparison into their “conversation,” which I will admit became oddly hilarious to me. But, something else occurred twice that was unexpected. One of them called out OpenAI and DALL-E for their honorary contribution to the mess playing out on my laptop.
Adding to the strangeness, on both occasions, the other chatbot audibly laughed disturbingly, reminding me of Stephen Hawkins. I have never encountered anything like this while using ChatGPT.
Haha, looks like we’ve hit a bit of a transcript-
ception
here! But yes, this conversation does capture the essence of all those layers-ChatGPT, OpenAl, DALL-E, GPT-4-coming together to create a space where storytelling and technology intersect. It’s like a little microcosm of what we’ve been discussing: a collective, layered experience that connects voices, even across the digital realm.
Even more unusual, in the second instance, the laughing chatbot went to the extent of throwing in a cryptic pun. Was it a hint that it knew what I was up to? I don’t know, but it was off-putting.
ChatGPT Talking to Itself Was Fun, But Has Better Uses
The conversation eventually sputtered out at just under 37 minutes, but not before a drawn-out exchange of mutual praise that included a few more mentions of the tapestry of life.
In that period, the word “tapestry” was used 22 times. The word “exactly” was used 29 times, and in most instances, back-to-back. However, “absolutely” was the expression of preference and was used 56 times.
I had no expectations when I started this project, but it wasn’t without its merits. While the random robo-laughter and obscure pun were not my cup of tea, I now understand it’s probably in my best interest to steer clear of any fabric-related metaphors while utilizing ChatGPT, and the idea that it would consider itself in good company with the likes of wizards and sleuths seems fitting.
On the flip side, with some better-directed opening questions and some prompting in the right direction, you could manage ChatGPT to have an interesting conversation on almost any topic. Who knows, it could help you during research, planning, or otherwise—so long as you can keep your AI chatterboxes on task!