Labs · Human or AI
Four LLM judges scored 12% on this test.
Your move.
One question a day. Two passages of fiction, read blind — pick the one that went through human editorial review, or tell two models apart by their fingerprints: cliché lexicons, simile density, sentence-rhythm flatness. Then copy your scorecard and go gloat.
Free · no account · record stays in your browser
Where this tool comes from
The question bank is generated from two corpora we own outright. Corpus one: our published fiction line — AI-drafted novels that passed structural validation, a scoring gate and human read-through before shipping (10 Chinese titles, 2026-07). Corpus two: the raw outputs of our nine-model continuation benchmark (2026-07) — same book, same anchor point, same instruction, same temperature, models from DeepSeek, Anthropic, Google, xAI, Moonshot, Zhipu, Alibaba and OpenAI. Model-vs-model rounds marked "same opening" pair two models' first-round continuations of the identical anchor.
The fingerprint highlights use the same v10.7 rule lexicon that powers our AI-Flavor Checkup (cliché phrases, narrated-attribution patterns, pseudo-precision), plus structural metrics: sentence-length dispersion, simile density, dialogue ratio. The "12%" anchor comes from our double-blind judge-reliability experiment (4 judges × 7 pairs × both orders, gold from real readers, 2026-07) — full write-up linked from the methods of our model leaderboard.
What it can't do
Getting one right doesn't make you a detector, and neither are we: a structurally perfect AI passage can pass every rule on this page. Our own gold-calibration work is exactly why we say "no silver bullet for human-ness" — the daily game trains your eye, it doesn't certify it.
There's no "human-written" round yet. Our published novels are AI-drafted with human editorial control, and we say so on every reveal. A true human-prose round waits for passages we have clean display rights to — community licensing is planned, and it will be labeled.
Today's passages are in Chinese: the measured corpus behind the game is Chinese-first. An English bank is planned.
No global accuracy counter yet — this page has no backend. Your opponent for now is the fixed 12% judge score, which is real, dated and sourced.
FAQ
Where do the questions come from, and how do they rotate?
Two copyright-clean sources, both ours: passages from our own published novels (AI-drafted, then run through structural validation, a scoring gate and human read-through), and raw continuation outputs from our nine-model benchmark (2026-07, same book, same anchor, same instruction, same temperature). A build script slices them into a 60-day bank; the day number picks the question deterministically, rolling over at midnight UTC+8. Past days stay playable as practice.
What's the "LLM judges scored 12%" line about?
A double-blind reliability experiment we ran in 2026-07: four heterogeneous LLM judges rated 7 pairs of character-card prose in both orders, against gold labels from real readers. They agreed with each other 86% of the time — and scored 2/16 = 12% against gold. Unanimously wrong: they read dense, evenly-polished detail as "human" when real readers read it as AI. That number is your opponent here.
Why does the question say "went through human editorial review" instead of "written by a human"?
Because our published novels are AI-drafted too — the honest difference is the pipeline: validation, scoring gates and human read-through versus raw, untouched model output. A true "human vs AI" round needs human prose we have clean rights to display; we're collecting community-licensed passages for a later season and will label it when it ships.
Where is my record stored? Is there a leaderboard?
In your browser's localStorage only — no account, nothing uploaded. There's no global accuracy stat yet (this page has no backend); the 12% judge score serves as the fixed benchmark to beat. A crowd accuracy line is on the roadmap.