When a language model writes, what is it actually doing?
Tokens, prediction, temperature, and the gap between fluent output and understanding. We start unplugged — guessing the next word ourselves — then open the machine and watch it do the same thing, visibly.
Text is chopped into pieces a model can count — often not where a human would cut.
The model only ever guesses the next token from what came before.
One dial slides output from predictable to surprising — and unstable.
Fluency is easy to mistake for understanding. ELIZA shows why.
Welcome & norms
Name the question. Remind everyone direct AI use is optional — each person chooses a participation pathway.
Unplugged prediction game
Show a sentence stem, collect likely next words. Notice how context, genre, and expectation shape guesses.
Tokenizer + Temperature
Show token chunks, then run one prompt at low and high temperature. Watch probability bars and greedy vs. sampled output.
ELIZA vs. frozen comparison text
Inspect the matched rule, pattern, and template. Compare visible rule-matching with frozen LLM-style text, while naming its incomplete provenance.
Investigation
Complete one Text Experiment Board section: a default test, a prompt variation, or an ELIZA comparison.
Debrief
Reflection frame — human, machine, system, ethics, pedagogy. Share one evidence-based claim and one question.
Participants can complete the whole session with only the visualizer, ELIZA, and frozen comparison examples — no logging into or prompting a live LLM. The comparison text should not be attributed to a named model because its original provenance was not preserved. Opting out of direct AI use never means opting out of the camp.