At every step, the AI has several possible next words to choose from. Two small dials tilt which one it picks — drag them and watch the choices open up or narrow down.
Narrated by a synthetic voice — not a recording
The harbour held its breath as the storm …
evidence marker · fixed
Whether the sentence is supported is a separate question. As you work the dials, watch whether anything on this page can reach this marker.
Try it — drag a dial and watch the possible words thin out or spread; re-roll to let the AI write a fresh sentence.
How each word gets chosen
The AI does not write a whole sentence and then show it to you. It picks one word at a time, adds it, and then works out a fresh set of choices for the word after that. The map above makes all the words it almost picked visible.
Some systems always pick the single most likely word. That makes the answer repeatable, but it does not make it any more truthful.
You’re reading the pocket version — drag with a thumb and tap re-roll; on a desk, the paths also bend toward your cursor.
Two dials sit under the map. Each steps in at a different point. Walk them one at a time ↓
Each step begins as a list of possible next words. The AI gives every word its own hunch for how likely it should be. A thicker line means the AI expects to pick that word more often.
Notice — the percentages beside the words always add up to 100%. Each line's thickness is just that word's share of one single choice.The first dial changes how safe or adventurous the AI is (engineers call it temperature). Turn it down and the AI plays it safe — the front-runner grows into one thick line and the rest shrink away. Turn it up and the AI gets adventurous — the gaps close and less-likely words get room to appear.
This dial only reshapes the list the AI already has — it changes how boldly one word is picked over another. It never looks anything up in a source.
Try it — drag the adventurousness dial from cold to hot in one slow pass; watch one line thicken into a trunk, then the thin lines crowd back in.Complete · A safer setting concentrates the choice; an adventurous one spreads it out.Next: use the other dial to remove words ↓
The second dial keeps only the most likely words and cuts the rest off the bottom of the list (engineers call it top-p). Watch the thin lines vanish outright rather than merely fade.
Try it — hold the adventurousness dial high and pull this one down; count the words that vanish outright rather than fade.Complete · This dial keeps the most likely words and removes the long list of unlikely ones.Next: let one surviving word become a sentence ↓
The tangerine line is one random pick from the words that survived the shortlist. Re-roll and a completely different sentence can appear from exactly the same setup.
Try it — re-roll five times without touching the dials; read the sentence it draws each time.Complete · Each pick lands on one word; the other words were still possible.You've seen the AI choose. Now ask what it cannot choose ↓
The AI can make a totally unsupported guess look very likely. Notice the cyan evidence marker: it never moved. These dials control how the wording is chosen. They cannot tell us whether a claim is actually supported by facts.
Try it — throw both dials to their extremes and re-roll; watch the EVIDENCE row while everything else changes.You've found the boundary. One line to take with you ↓
draw · the storm broke → over → the pier