Because It Speaks in Words

There is a profound chasm between the act of knowing a fact and the experience of truly understanding it.

We have all encountered those rare intervals where a truth, previously learned as a dry piece of information, suddenly resonates. It is as if the concept expands within the mind, revealing a vast landscape of nuance that was previously invisible.

Consider a night when the Moon, Venus, and Jupiter align. Usually, we perceive them simply as $\text{dots} + \text{crescent}$. However, when framed by a twilight sky and silver clouds, the perspective shifts. You realize that you are gazing upon the exact same celestial arrangement witnessed by:

• Your own great-grandparents.
• Julius Caesar.
• Aristotle.

Despite our modern astronomical data, the visual experience remains unchanged; no human eye has ever seen more than what you see in that moment. This realization mirrors our current collective awakening to the potency of words—not merely as tools for the speaker, but as catalysts for the listener.

The Architecture of Narrative

Eugene Wigner once explored the "unreasonable effectiveness of mathematics" in the natural sciences. He noted that advanced math predicts nature with an eerie precision, suggesting that mathematics might be the $\text{true language}$ of the universe.

$\text{Nature} \approx \text{Mathematical Logic}$

Yet, humans are not born with an intuitive grasp of $\text{calculus}$ or $\text{linear algebra}$. Instead, we are born with an innate, driving hunger for stories—tales of ourselves, our neighbors, and the cosmos.

Our Cognitive Wiring

Our brains are not objective processors; they are biased toward narrative. We are prone to:

1. Preferring stories that "rhyme" (possess internal symmetry) over those that are disjointed.
2. Trusting the person speaking to us.

This second bias is perilous in the modern era, as we are submerged in a sea of words designed specifically to be believed.

Stories serve as our primary mechanism for transmitting data, issuing warnings, ostracizing the deviant, or celebrating the hero. We construct narratives to explain our successes or failures, and we often moralize disease because a story of "sin and punishment" is psychologically easier to digest than the cold, $\text{strikethrough}$ randomness chaos of a stochastic world.

Ultimately, the only firsthand truth we possess is the collection of stories we tell ourselves. We communicate these via gestures and actions, but primarily through words. This makes language the most influential force in human civilization—the glue of our social fabric. We use words to invoke authority, laud the brilliant, or dismiss the foolish. Even the "jewel" of mathematics is transmitted and taught through the medium of language.

Image: The allure of the voice, reminiscent of Waterhouse's themes.

The Digital Siren Song

While disinformation is not new, we have entered an era of unprecedented auditory and visual noise. The Ancient Greeks feared the Sirens' entrancing songs; today, digital sirens wail in our pockets and on our screens every second.

We have reached a "step function" in communication. This leap isn't due to a sudden evolution in human eloquence, but because we have created an interlocutor that is fundamentally different from us.

It is a stochastic parrot: an entity that mimics the patterns of our speech without possessing a shred of the understanding behind it.

Despite this, we are predisposed to learn from it and believe it, regardless of its factual accuracy.

Intelligence vs. Mimicry

For years, various machine learning (ML) tools have existed. For example, Support Vector Machines (SVMs) and Random Forests are exceptional at extracting signals from noise.

I recall when Apple provided a drag-and-drop interface for developers to create simple recommendation models. These were powerful, yet they never captured the public imagination. Why did Large Language Models (LLMs) trigger a global frenzy while other ML techniques did not?

It isn't necessarily because LLMs are "better" algorithms in a vacuum. It is because they stopped speaking in vectors and tensors and started speaking in words.

# Traditional ML output: A vector of probabilities
output = [0.12, 0.88, 0.04, 0.91] 

# LLM output: A narrative
output = "I believe the answer to your question is..."

Words are our exclusive domain. We call LLMs "intelligent" and fear them because they sound like another mind. Throughout history, every new communication medium—from pamphlets to emails—has been used to seek out other humans and share stories. Those stories usually correlated with some version of a human truth.

Now, we face a new reality:

• Language that sounds genuine.
• Statistical precision in mimicry.
• A mind or experience to convey.

We are captivated not by the machine's intelligence, but by our own deep-seated belief in the power of the word.