Mark Levy/Writing

A pixel-art starship at warp, seen from above and behind, stars streaking past its glowing nacelles. Ahead of the bow, a small amber blank waits to be filled in.

[ An interactive explainer in three decks ]

The Word Warp Drive

How a large language model actually works, one guessing game at a time. Told aboard the USS Plausible.

Q
Ah. Captain. I've brought you aboard the USS Plausible, a ship that runs on exactly one trick: it guesses the next word. Three decks, seventeen sections, seventeen hand-built demos you can run, and at the bottom of each deck I put you on trial. Five charges, three shield levels, and you always come back; I'm cruel, not wasteful. No math degree required. Your progress is kept in this browser and nowhere else, which is more privacy than the Continuum ever gave me.

Every impressive thing these models do is one act, repeated: given some text, predict a plausible next word. The chat window, the personality, the tools, the agents: all ship, built around that core. The decks go in that order. Deck 1 is the core itself. Deck 2 is the bridge, the plumbing that turns it into a chat. Deck 3 is away missions, the loop that turns a chat into an agent, and it's where the vocabulary of current AI development lives.

Start on Deck 1 unless you've served before. Each deck stands on the one before it.

Q's termsI will compare this ship to car engines, Starfleet Academy, a holodeck detective, and a great many things from your species' television. Purists will note that a Galaxy-class starship has no carburetor. Correct. I am omnipotent, not consistent: the holodeck can render whatever scene gets the idea across, and I have watched all of your cinema, twice, out of boredom.

[ Deck 1 of 3 ]

Engineering: the core itself

Words become coordinates, a trillion settings, an academy that runs on fill-in-the-blank, sensors, the loop, and why fabrication is the core working as designed.

Stellar Cartography·The Positronic Net·The Academy·The Sensor Array·The Causality Loop·The Holodeck·Tribunal: Lore

Not yet boardedGo to Deck 1

[ Deck 2 of 3 ]

The Bridge: from question to answer

How a question becomes the start of a completion, what inference is, what "thinking" and "effort" actually buy, and how a picture rides the same loop.

The Deerstalker·Utopia Planitia·The Ready Room·Warp Factor·The Viewscreen·Tribunal: Moriarty

Not yet boardedGo to Deck 2

[ Deck 3 of 3 ]

Away Missions: the agent

Agents, tools and MCP, standing orders, away teams, safety protocols, evals, and context engineering, as the ship built around the one core.

Make It So·Standing Orders·Ship's Systems·Away Teams·Safety Protocols·The Briefing Room·Tribunal: the Borg

Not yet boardedGo to Deck 3

Standing orders

Each section is one idea and one demo. The demos are hand-built with made-up numbers; they show the mechanism, not real model output. On a wide screen a map of the deck sits on the left with a marker for where you are; on a phone the same controls are at the bottom. Every section ends with a captain's log entry, the thing worth carrying out, and a note on where the same idea shows up back on Earth, in the tools you use every day. Each deck ends with the ship's manifest of the words you now own, and then Q's tribunal: five charges about the sections above, three shield levels. A wrong answer drops the shields and names the section to revisit; lose them all and Q snaps his fingers and you're back at the door. The turbolift is never locked. Q just judges you.

The mental model follows John Mount's "A Simplified Mental Model of LLMs"; the teaching style, run the simulation instead of reading the bullet list, follows Laurentiu Gabriel's "How I Use LLMs to Learn". Deck 3 leans on Anthropic's writing on agents, context engineering, and evals. The starship framing is an affectionate homage to a certain 1987 television series, with no affiliation; the ship and its name are ours.

All writing · Back to marklevy.info