The Daily Demystify
One AI mystery a day — the same puzzle for everyone, Wordle-style. Each day serves a plausible-sounding belief about how AI works, three or four answer choices, and exactly one true one. Commit to an answer and the moat kicks in: instead of a slogan, it reveals the mechanism-level truth (why temperature 0 isn't perfectly deterministic, why a model miscounts letters, why it "hallucinates") and shows how often the tempting wrong answer fools the crowd. Runs entirely in your browser (0 uploads, works offline).
How to use it
- Read today's myth — the plausible-but-wrong belief at the top of the card.
- Pick the option you think is actually true. You get one commit per puzzle.
- See the verdict, then open the X-ray to read the mechanism-level truth and the crowd's answer split.
- Use Previous and Next mystery to browse the rotating bank, and copy or share your result grid.
What this clears up (the fundamentals)
- Plausible is not true — a fluent, confident AI answer runs on the same next-token machinery as a correct one. The model optimizes for what sounds right, not for ground truth.
- The reveal is about you, not the model — seeing that a large slice of people also picked the tempting wrong answer demystifies your own overconfidence, not just the AI.
- Mechanisms beat vibes — "temperature 0 is greedy decoding, but batching can still shift output" is a fact you can act on; "AI is unpredictable" is a mood.
- Same puzzle for everyone — rotation is deterministic by the day index, so the daily item is identical worldwide. No personalization, no tracking, no server call at play time.
Where it's used
A free, top-of-funnel daily habit — the homepage strip, the Playgrounds gallery, an opt-in daily nudge. It's a miatz build-lab concept playable: play the daily here, then learn to build the rotating item bank, the deterministic grader, and the crowd-reveal yourself.
FAQ
How is "today's" puzzle chosen?
By a day index — days since a fixed epoch — modulo the size of the item bank. That makes the rotation deterministic and identical for every visitor on the same day, with no server picking it for you.
Is the crowd answer-distribution real?
No — it's an illustrative, deterministic split computed offline from the day index, because the tool makes zero network calls. It's there to make the point (the plausible-wrong answer is tempting), not to report live global stats.
Why is a wrong answer's rationale still shown?
Because the reveal is the product. Whether you're right or wrong, the panel names the mechanism — the training cutoff, the context window, the token boundary — so you leave understanding why, not just whether.
Can I go back to earlier puzzles?
Yes. Previous and Next mystery walk the rotating bank so you can play more than one in a sitting. Nothing you do is stored off your device.
Is anything uploaded?
No. The item bank, the grading, and the reveal all run in your browser — nothing is transmitted, stored, or logged. Turn off your Wi-Fi and it still works.
Limits
A teaching widget with a small, hand-written item bank and an illustrative crowd split — it's a 60-second daily habit, not a benchmark or a live poll. The myths it busts are common ones; the mechanisms it names (tokens, context windows, training cutoffs, greedy decoding) are the real, load-bearing concepts.
Related
Part of the miatz playgrounds. Explore the rest from the playground catalog.
Bookmark this page (Ctrl+D, or ⌘D on Mac) — it works offline the next time you need it.