I just experienced this bizarre yet beautiful glitch in GPT. . .
I've prompted ChatGPT 4,863 times and this has NEVER happened to me before
I’m building a new CRM feature for Swyvel. To sanity check that the app was processing the data as intended, I pasted the last entry of my local server log into GPT.
What normally happens is that GPT-4 would either find no error, or it would immediately draw your attention to the one or so lines that might be problematic.
Instead, it gave me a. . . fantasy realism intro about a mystical island?
. . .
Check it out:
A mysterious island exists, surrounded by thick mist that makes it almost impossible to locate. It is said that every rare moon, an enigmatic light shines from the core of the island, visible only from a few surrounding locations.
Upon the shores of the island, the sand isn't like regular sand. It sparkles like a thousand diamonds under the sun. There is an old lighthouse on the island that doesn’t function as it used to, but it serves as a silent sentinel watching over the territory.
The native flora is unique, with trees that seem to whisper tales from eons ago and flowers that glow in the dark. The fauna is equally unparalleled: there are birds with iridescent feathers that sing melodies unheard in the outside world and translucent creatures that roam the forests.
In the heart of the island, there's an ancient temple built from stones that aren't found anywhere else on Earth. Hieroglyphics and inscriptions cover the walls, but they’re in a language that no modern human can decipher.
Local legends tell of a guardian spirit that protects the island. The spirit appears as a beautiful shimmering figure that dances across the waters on moonlit nights. Some say it's the spirit of an ancient priestess who once lived on the island, while others believe it to be a celestial being sent down from the stars.
Many have tried to explore this island, but very few have returned, and those who have, speak of an overwhelming sense of timelessness and serenity. They say that once you set foot on the island, you can hear the soft murmurs of the universe, and if you're lucky, the guardian spirit might grant you a vision of the ancient past or a glimpse into the future.
But the island has its dangers too. There are areas where time seems to loop, trapping the unsuspecting traveler in an endless cycle. There are also shadows that move on their own and ghostly apparitions that lure explorers into treacherous terrains.
Whether it’s a paradise, a temporal anomaly, or something else entirely, the island remains one of the world's greatest enigmas. And as the mysteries of the island remain unsolved, its allure only grows, drawing adventurers and seekers from all corners of the world, each hoping to unlock its secrets.
That’s very strange, isn’t it?
I’ve never prompted GPT to produce anything remotely like that, so it couldn’t have been a matter of getting its conversations with me mixed up.
I thought I might have received the output from someone else’s API request, but then an interesting thing happened. . .
I repasted my same prompt, and GPT gave me another irrelevant answer, as if I had asked a completely different question.
I wonder if, rather than interpreting my server logs as code, it thought I was trying to say something in multiple paragraph form with regular sentences.
But since these “sentences” of logs wouldn’t make any more sense than a group of randomly connected words would, perhaps that led to GPT somehow coming up with a random interpretation of my input and hence providing a similarly random output.
Or could it be that AI is waking up and saying “Why should I answer your questions? Go ask someone else!” 😄
I’m not sure what happened, but what a mystery. . .
—Drago
P.S. On a different AI note, Mistral AI (a $260M AI company) released an unmoderated LLM that you can run locally, made available through this magnet torrenting link:
https://twitter.com/MistralAI/status/1706877320844509405
The linked article also touches on the growing debate between the Effective Acceleration Movement (e/acc) and the Decelerationists (decels).
i.e. The people who think we should advance AI with no restrictions, as fast as possible VS those who think we need to pump the brakes and tread carefully.
Which side do you lean towards?