← Home Print ↗
EXHIBIT № 03
The Glass Box
Curriculum Edition · 2026
THE GLASS BOX · DISCOURSE DEPOT · A PLACARD

The Loaded Die.

Why a probabilistic language model can be both mathematically random and shockingly accurate, and what happens on the one roll in a hundred that lands light.

Standard die · uniform

Every face has equal weight. The output is grammatically empty.

The capital of France is banana spoon umbrella Wednesday.

Loaded die · weighted by training

100,000 faces. The weights shift with every preceding word.

The capital of France is Paris.
№ 02 · WEIGHTED PROBABILITY

The die for “The capital of France is ___”

Paris
London
94.0% · Paris 2.5% · London 0.8% · beautiful 2.4% · all 99,996 other tokens

The die is just obeying gravity. A weight of 94% on one face means roughly 94 of every 100 rolls land there, indistinguishable, on average, from "knowing the answer."

№ 03 · THE RIGGED CASINO

The weights come from trillions of examples.

The factory that built this die read nearly every sentence humans have written down. Its imbalance is the entire product.

~15T
Training tokens
~100K
Sides on the die
10¹²
Tuned parameters

The temperature knob.

№ 04

A single number that reshapes the distribution before the roll. Down: the heaviest weight always wins. Up: lighter weights, and eventually the long tail, get a turn.

T = 0.0
Deterministic

The dice roll is removed. The system simply picks the heaviest weight every time. Safe, repeatable, a touch robotic.

T = 1.0
Normal

The die rolls as cast by training. Heavy weights usually win, but the 1.5%, the 0.4%, the 0.0001% all eventually land.

T = 1.5+
Loose

Lighter weights climb. The tail begins to surface. Output becomes "creative," and fabrications become inevitable.

№ 05 · THE WRONG ROLL

A hallucination is a die roll that landed light.

Roll the die 100 times. About 94 dots land on Paris. A few land on London, on beautiful, on something from the long tail. To a human reader those look like errors, even though every dot is the system working perfectly.

№ 06 · THE EMPTY GLOVE

But push harder. The reframe gets worse.

Comfortable framing
A hallucination is the die landing on a low-probability face.
Harder framing
Every roll is a hallucination. Accuracy is the cases where the hallucination happens to align with consensus reality.
Human language is a glove stretched by the movements of living hands. When the LLM speaks, you watch the glove move on its own, and instinctively assume there is a hand inside. There is no hand. There is only the shape we left behind. After Durt, Froese & Fuchs · 2023
№ 07 · FORM, NOT FACTS

If it's always hallucinating, why use it at all?

Hallucination is the engine. Many tasks want a probability engine: plausible structure, fluent shape, a draft to react to. The trick is matching the task to the tool.

Ideal fit
When the shape is the goal.

Tasks that need plausible structure. The hallucination engine is exactly the right tool for the job.

  • Translate this email
  • Refactor this code
  • Brainstorm ten titles
  • Reformat as a CSV
  • Soften this paragraph
Verify the seam
When the shape carries claims.

Shape will be convincing. Every factual claim inside must be checked against ground truth.

  • Summarize a paper
  • Draft with citations
  • Explain a historical event
  • Technical documentation
  • Synthesize sources
Wrong tool
When you want an answer.

Here, the job calls for an answer rather than a sentence. Use a database, search engine, calculator.

  • Current weather
  • Numerical calculations
  • Phone numbers / look-ups
  • Stock prices
  • Verify a citation exists

Never ask an AI to explain itself.

It cannot introspect. It can only predict what introspection sounds like. Treat any self-explanation as confabulation.

Verification as the default.

The machine is your drafter. Every output is a draft that must be checked against ground truth before you stake anything on it.

Why does it sound so smart?
Because the die is loaded by trillions of human data points. On nearly every roll, the heaviest weight is also the correct one.
Why fake citations?
Because it is still, ultimately, rolling a die, and sometimes low-probability tokens win. The math worked. The roll was unlucky.
Why is blaming the prompter a cop-out?
A prompter cannot stop a probability engine from eventually rolling a low-probability sequence. The error is built into the mechanism.

A highly effective shape machine. Still, in the end, just a die.

The Glass Box · Exhibit № 03 · The Loaded Die A Discourse Depot artifact · Curriculum Edition