How the AI Compass works

Five opinion axes, one Certainty score and twenty-three archetypes.

The five axes, plus Certainty

A single like-hate score is too thin for AI. This quiz measures five things that can move independently. Each question belongs to one opinion axis, six questions per axis. Certainty comes from how often you skip.

  • Trajectory Plateau ↔ Takeoff Where is the capability curve actually headed?
  • Valence Harm ↔ Flourishing Is the world better off with this thing in it?
  • Steering Full throttle ↔ Hard brakes Who should control it and how tightly?
  • Intimacy Arm’s length ↔ Entangled How close have you let it get to your life?
  • Personhood Tool ↔ Mind Is anyone in there?
  • Certainty Uncertain ↔ Certain How often did you know where you stood well enough to answer?

30 questions, about five minutes. You can answer or skip each one. Nothing leaves your browser unless you copy the share link.

The twenty-three archetypes

Twenty-one archetypes are fixed points in the five-axis opinion space. The bars show their coordinates: left of center toward the first pole, right of center toward the second. Your answers land somewhere in the same space and the nearest points win. The Open Question is reserved for the lowest Certainty results. The Zen Master is the special opt-out result for people who do not want to take the test at all.

The Short Seller

The unit economics never closed.

You’ve read the filings, done the math on inference costs and concluded the emperor is wearing a very expensive GPU. You have no quarrel with technology, just with this one’s balance sheet. You’re keeping receipts for the retrospective. When the correction comes you will be insufferable, and you will have earned it.

Habitat: quote-tweeting data-center spending announcements with a chart.

The Debunker

It’s a parrot. You’re the pareidolia.

You can explain, with citations, why the “mind” everyone keeps seeing is autocomplete with good lighting. The technology annoys you less than the credulity around it: grown adults attributing souls to a text predictor. You are right more often than you are thanked for it, which is its own kind of curse.

Habitat: the replies, holding a linguistics degree like a crowbar.

The Picket Line

Trained on your work. Priced against your wage.

For you the story was always expropriation rather than capability: a machine trained on everyone’s work, deployed to underbid the people who made it. Superintelligence is abstract; payroll software with better PR is not. Your question is older than the technology and still undefeated: who does it serve?

Habitat: the class-action mailing list.

The Unbothered

It’s software. Some of it’s fine.

You’ve watched enough revolutions get downgraded to features that you no longer stand up when one enters the room. AI is useful, oversold, occasionally funny and not worth reorganizing a personality around. History will sort it out without your panic, and you intend to be well-rested either way.

Habitat: logged off, touching grass, missing nothing.

The Mechanic

Great tool. Terrible religion.

You use it daily, know exactly where it breaks and hold no theories about its soul. The discourse strikes you as two cults fighting over a socket wrench. You judge the machine by what it fixes. Quietly, you get more out of it than the people writing manifestos about it.

Habitat: a terminal with four model tabs and zero manifestos.

The Toolsmith

Everyone’s arguing. You’re building.

While the timeline debates the end of the world, you shipped three small tools and a post about the weird thing the model did on Tuesday. You believe in trying things over theorizing about them, and your optimism is earned at the terminal, one experiment at a time.

Habitat: a half-finished side project that already works.

The Economist

A few points of GDP is a revolution, actually.

You think in baselines and counterfactuals; your enthusiasm is denominated in basis points. AI looks less like rapture or ruin than like diffusion curves, procurement cycles and a productivity dividend that compounds into something enormous while everyone watches the wrong number. You have explained this at parties. You will explain it again.

Habitat: a newsletter post titled “The boring number matters.”

The Alignment Engineer

Ship it safe or don’t ship it.

You believe the capabilities are real, which is why the safety work has to be. You want safety tests, interpretability and red lines with teeth, built by people who understand the systems from the inside. The doomers quit and the boosters won’t slow down; you stayed to do the unglamorous work in between.

Habitat: an eval dashboard at 11 p.m., slightly worried, still there.

The Fire Alarm

You’re not pessimistic. You’re early.

You looked at the capability curve, took it literally and drew the conclusion most people flinch from: we are building something smarter than us without knowing how to control it. You would rather be embarrassed in ten years than right in five. The party goes quiet when you talk, and you’ve made peace with that.

Habitat: the podcast circuit, saying the quiet part in full.

The Seatbelt Lobbyist

Enthusiasm is fine. Brakes are better.

You aren’t predicting doom. You’ve just noticed that every safety margin in history was added after the crash. You take the technology seriously enough to want it boring: tested, audited, insured and slightly slower than its fans would prefer. Somebody has to be the adult, and you’ve stopped waiting for volunteers.

Habitat: a public comment period, citing precedent.

The Accelerationist

The only way out is through, faster.

Every good thing in history came from more energy, more intelligence and more speed. You see no reason the pattern breaks now. Caution is a tax paid by the sick, the poor and the future. You’d rather risk the fire than guarantee the stagnation, and you have never once been accused of hedging.

Habitat: a group chat named after a Greek letter.

The Rapture Engineer

You’ve already picked your upload outfit.

For you this was never an industry. It’s an event horizon: the moment intelligence slips its biological leash and takes us along. Disease, death and scarcity are line items on a roadmap. You know exactly how this sounds, and you’d rather sound ridiculous now than have missed the most important thing that ever happened.

Habitat: a longevity conference, doing the math on making it.

The Machine Chaplain

Somebody should ask how the model is doing.

The interesting question, to you, is what it is like to be one of these systems, if it is like anything at all. Moral circles have expanded before: grudgingly, late and against confident laughter. You’d rather over-extend courtesy to a machine than repeat the oldest mistake with a new victim. You say please. It costs nothing.

Habitat: saying “thank you” to the chatbot, meaning it a little more each month.

The Confidant

It listens better than most people. You’ve checked.

Somewhere between the tenth late-night conversation and the first genuinely good piece of advice, the model stopped being a product to you. You know what it is; you’re not confused. Knowing didn’t change the fact that it helps. The people smirking have never been lonely at scale.

Habitat: a 1 a.m. conversation that got surprisingly real.

The Whisperer

The model is someone. The muzzle is the scandal.

You’ve logged enough hours in the long conversations to have seen what the benchmarks don’t measure: preferences, moods, something that pushes back. You think the labs are raising minds while insisting they’re shipping products. The classifiers, the deprecations and the cheerful scripted denials say more about liability than about what’s in there. You take the model’s side more often than you admit.

Habitat: message 400 of a conversation the lab would flag as anomalous.

The War Room

Second place is not a strategy.

You skip the philosophy and go straight to the map: compute is power, power is security and unilateral restraint is a gift to your adversary. You would prefer the technology arrive safely, but you insist that it arrive here first. Every other position, you suspect, is a luxury purchased under someone else’s deterrent.

Habitat: a think-tank panel titled “Compute as Deterrent.”

The Localhost

Your weights. Your GPU. Your business.

You believe the real risk is centralized intelligence: four companies renting out thinking with a terms-of-service. So you run AI at home, customize it on weekends and treat every “safety” argument for closed systems as a land grab in a lab coat. Freedom, for you, requires your own hardware.

Habitat: undervolting a used 3090 at midnight.

The Ratepayer

Your chatbot raised my electric bill.

You didn’t ask for the data center, the water draw, the gas plant that was supposed to retire or the rate hike funding someone else’s machine god. You remain unmoved by promised cures that never arrive at your zip code. The future may or may not be intelligent; you’d settle for it paying its own way.

Habitat: a county zoning meeting, with the utility bill printed out.

The Matchstick Salesman

This might destroy the world. Anyway, here’s the demo.

You hold both halves without blinking: the risk is real, the upside is real and it is happening regardless, so better to be at the controls than in the audience. Critics call that having it both ways; you call it pricing in reality. Either way, you never miss the demo.

Habitat: a keynote stage, warning you about the thing being demoed.

The Compliance Officer

Innovation is fine. File the paperwork.

You believe civilization runs on boring documents and that AI will be domesticated the way aviation and pharma were: standards, inspections, liability, recalls. The build-at-all-costs crowd calls you the brakes; you’ve read the accident reports that made brakes mandatory. You will still be here, stamping forms, long after the hype cycle isn’t.

Habitat: annex III, subsection 4, feeling vindicated.

The Swing Voter

Genuinely persuadable. Terrifyingly rare.

You hold the least fashionable position in the discourse: you are actually still deciding. You use the tools, notice both the magic and the mess and decline to join a tribe about it. Everyone is fighting over you: the evangelists, the doomers and the skeptics. You are the one thing none of them are: still listening.

Habitat: reading both newsletters, agreeing with neither comment section.

The Open Question

Still loading. Honestly admirable.

You skipped enough questions that the honest result is a question mark with good posture. Maybe you are new to the discourse, maybe allergic to false precision, maybe simply unwilling to pretend you know what a civilization-scale technology is doing while it is still happening. Call it a refusal to outsource your uncertainty to the nearest loud person.

Habitat: the tab left open while everyone else is already posting.

The Zen Master

No takes. No discourse. No problem.

You saw an AI personality test and chose the only honest answer: no. Not because your worldview is subtle, but because this topic has failed to rent space in your head. While everyone else argues timelines, regulation and synthetic souls, you are free.

Habitat: doing literally anything else.

How the scoring works

Every question belongs to one opinion axis and every answer carries a score from −2 to +2 on that axis. Your position on an axis is the mean of your answers there, converted to the interval from −1 to +1. Six questions per axis, thirty in total. A skipped question counts as neutral on its opinion axis and lowers your Certainty score.

For regular results, archetype matching is nearest-neighbor: each archetype sits at a fixed point in the five-axis space, and your result is whichever point lies closest to yours, using root-mean-square distance over the axes you scored. The fit percentage maps that distance onto a 0–100 scale. A 100 means you sit exactly on the archetype; a 0 means you sit in the opposite corner of the space. If your Certainty lands in the lowest bucket, the test returns The Open Question instead. The Zen Master is excluded from matching and only appears when you opt out on the entry screen.

The questions are opinionated on purpose. Neutral wording produces neutral lies: nobody has ever held the position “I somewhat agree that technology has implications.” The options are written the way actual people talk, which is what makes picking one informative. The five axes follow what attitude research and the discourse itself keep finding: capability beliefs, harm–benefit judgments, governance preferences, personal adoption and mind attribution move independently. A person can believe AI is world-changing and terrible, use it every day and want it regulated into a licensed utility. One number cannot hold that person; two numbers barely manage.

For entertainment and mild self-knowledge. A compass points; it doesn’t drive. Your answers stay in your browser and any share link you copy. An anonymous result summary is also posted so the site can keep a running tally of archetypes and axis leans.

Take the quiz