Reads like an AI. Calculates like a proof.
The language model only reads your question and picks the principles. Every number is worked out and checked by Mystery's own engine.
Filled marks: training data. Hollow marks: unseen test data. The background colour is the network's guess everywhere.
Line thickness shows each weight's size: blue = positive, red = negative.
What is happening, step by step
Start with forall to test a claim, or find to search for an example:
forall n in 1..10^6: isPrime(n^2 + n + 41)forall n in 4..10^7 step 2: exists p in 2..n/2 where isPrime(p): isPrime(n - p)find a in 1..100, b in a..100, c in b..150: a^2 + b^2 == c^2
Add where … before the colon to skip cases, and track … at the end to chart a value. Inside you can use + − * / // % ^ !, == != < <= > >=, and / or / not, and exists / forall / count / sum / product / min / max. Functions:
Try a famous one
How the referee works
1 · Choose data
2 · Choose competing theories
1 · Choose a challenge
How the loop works
Your code is checked on this phone. Only when you press "Fix it" are the few lines needed sent to the AI, and secret keys are removed first.