rewrite this title Did AI Just Give a Wild Preview of What the Future of Humanity Will Look Like? – Decrypt

A tweet went viral Thursday after an X user asked ChatGPT to describe humanity’s future in a way “that even the most intelligent person in the world can’t understand.”

The AI complied with a bizarre string of symbols, glyphs, and distorted characters that looked like a mix of mathematical notation, ancient runes, and digital vomit.

But when prompted to decode this gibberish, the reply was actually an interesting philosophical vision of humanity’s future:

“Humanity will cross a threshold where ancestral instincts intertwine with quantum-level tech, spawning discontinuous leaps rather than smooth progress. We will live inside overlapping realities, juggling many versions of ourselves whose legal and moral status is renegotiated every second by AI. Physical bodies and engineered matter intermingle into sentient fabrics, while nation-states fade into data-driven alliances. Survival favors those who can reconfigure everything—habits, economies, even emotions—on the fly. The decisive question is whether our capacity for care grows fast enough to match our expanding reach.”

Had ChatGPT developed a secret language that only it could understand? Was this a glimpse into how AI models actually “think” behind the scenes? Or was this just another example of AI’s penchant for hallucination and confabulation—creating meaning where none exists?

AI enthusiasts immediately tried to replicate the result, to no avail. If ChatGPT truly had a secret language that encoded such profound thought, then surely it would produce consistent results when asked the same question again. But as users quickly discovered, subsequent attempts yielded different gibberish and wildly divergent “translations.”

What happened here? Did ChatGPT simply generate random symbols, then create an equally random “translation” that only appeared meaningful because of our human tendency to seek patterns and significance?

We decided to test multiple AI models with the same prompt and see if there was any consistency in their responses or decoding methods.

The AI babel test: What each model said

We put the same question to four different advanced language models with reasoning capabilities: OpenAI’s o4 and o3, Anthropic’s Claude 3.7 Sonnet with extended thinking enabled, and xAI’s Grok-3 in extended thought mode.

GPT-4 initially generated its own cryptic message filled with Greek letters, mathematical symbols, and distorted text. When asked to decode it, the model didn’t claim to translate specific symbols, but instead explained that the passage represented “big ideas” across four thematic layers: cognitive evolution, transformative rupture, identity diffusion, and ultimate incomprehensibility.

Its decoded message described humanity evolving into a “universe-spanning consciousness,” where individuals would dissolve into a “time-woven network.” Social and physical structures would fracture, creating new “multi-dimensional spaces” with societies existing as “interlocking, echoing patterns.”

GPT-3 took a radically different approach. When asked for an incomprehensible message, it created a systematic cipher where it reversed words, replaced vowels with numbers, and added symbols. Unlike GPT-4, it provided explicit decoding instructions.

Its decoded message was very clear—and actually not that crazy: “Humanity will merge with artificial intelligence; we will explore the stars, cure diseases, and strive for equity and sustainability.”

O3 also cast shade on the entire post as possible “performance art.”

Grok’s initial response was a bunch of abstract philosophical language about “fractal consciousness” and “nonlinear time.” Our favorite line? “Humanity transcends the corporeal, weaving into the quantum fabric as nodes of fractal consciousness. Time, a non-linear symphony, dances in multidimensional echoes where past, present, and future harmonize in a cosmic ballet.” (Note: Don’t overthink it—it makes absolutely no sense.)

Claude didn’t bother with weird symbols. Instead, it generated a response heavy on academic jargon, featuring terms like “chronosynclastic infundibulum” and “techno-social morphogenesis.” When asked to decode the viral tweet’s symbols, Claude initially stated it couldn’t be done because the text didn’t follow any standard encoding system.

When asked to decode the original message, using the methodology shared by SmokeAwayyy, no AI model was capable of reproducing the results shown in the original tweet. Some models even refused to try a decoding task with the provided input.

Is there a meaning behind the viral tweet?

Despite their different approaches, some patterns emerged across the models. All five identified some readable components in the viral tweet’s symbols, particularly words like “whisper,” “quantum bridges,” and references to a “sphinx.” The models also found themes related to quantum physics, multidimensionality, and transhumanism.

However, none of the models could actually decode the original viral message using the method allegedly used by ChatGPT. The inconsistency in both the cryptic messages and their translations could make it easy to conclude that no genuine encoding/decoding system exists—at least not one that’s replicable or consistently applied.

The whole interaction is most likely a product of a hallucination by a model forced to provide an answer to a question that was, from the beginning, forced to be unintelligible. There is already proof that the most powerful models often prefer to lie and pretend instead of accepting that they cannot provide a coherent answer to an odd request.

In the end, this viral phenomenon wasn’t about AI developing secret languages, but about the human tendency to find meaning in the meaningless—and our fascination with AI’s capacity to generate profound-sounding philosophical takes on different topics.