Distance of Thinking

Summary

“Generative AI doesn't think. It predicts.” The piece shows not the answer but how the AI arrives at it: the path a visitor flies picks the model's next word, and that path becomes the answer.

6

next-token candidates at every step

the model's own probabilities

11

stages, from prompt to residual memory

4 days

to the prototype, in 22 commits

Personal work – concept, interaction design, visuals, frontend, server and hardware controller

←→

See all1 / 10

Overview

The piece treats AI not as a thinking being but as a system that keeps calculating the probability of the next token; a hallucination is a probability that converged in the wrong place, not a lie. Visitors pilot a ship through a tunnel of probabilities the model actually computed, and each word they pass is committed to the answer.

How it works

1. Tokens – the visitor's sentence is split with the real Llama 3 tokenizer.
2. Meaning space – each token's 4,096-dimension embedding is projected into 3D, and tokens that land close are joined by lines.
3. Next-token candidates – the model is asked for one token at a time, returning its top six candidates and their probabilities. The likeliest sits straight ahead; less likely ones form the outer rings of a tunnel.
4. Steering – a tilt sensor (MPU6050) picks a branch and a rotary dial sets temperature, which widens the field on screen and is sent to the model itself.
5. Hallucination – straying into unlikely branches raises instability by itself, and misspelled "ghost" words appear.

Key decisions

1. 01No separate answer – the flight path is the answer

   What the piece shows had to be what actually happens. Only by intervening in sampling is the answer literally made by the visitor's choices.

   →Each token passed is committed, one by one, into a sentence. Flying hands-off follows the likeliest path.

2. 02Real probabilities, not imitated ones

   Cloud models don't expose real probabilities, so the piece couldn't claim to show the model's internal state.

   →Local Ollama logprobs supply the real candidates and probabilities, and the piece runs offline in the gallery.

3. 03Pick the model by language

   EXAONE writes Korean well but looped on English prompts.

   →Korean prompts are answered by EXAONE 3.5, English ones by Llama 3.

4. 04Drop the shake input; tilt only steers

   If shaking the sensor caused hallucinations, the cause would be the visitor's hand. A hallucination had to come from unlikely choices.

   →Instability builds only when flying into low-probability tokens.

5. 05No colour – black and white only

   The moment warning colours appear, it looks like any 'AI UI'.

   →Meaning is carried by density, opacity, size and motion alone. The reference is Ryoji Ikeda's data-verse.

Built for the gallery

• Exhibition mode – operator UI hides, and after 75 seconds without input an attract loop takes over.
• Warm-up – a button preloads the models before visitors arrive; if generation fails, a local fallback keeps the visuals running.
• Network – Web Serial needs a secure page, so instead of a cloud deploy it runs on one HTTPS address on a private network.

Reflection

• It persuades only if what it shows is true. Choosing real values over imitated ones made the concept and the build say the same thing.
• Not done – the physical sculpture wasn't made, the Arduino controller wasn't tested on real hardware, and long-run stability for exhibition remains.