Mission · with Hunch

Hunch: Train, Then Test

AI learns from examples people label. Bad labels teach bad lessons. Train Hunch, then test it on signals it has never seen.

Steps

  1. Labels are lessons

    When you label training examples, you are teaching. If a label is wrong, Hunch learns the wrong thing.

  2. Features

    Hunch only looks at the features you switch on. Watching useless features adds noise.

  3. Held-out test

    The real test uses examples Hunch never trained on. That is how you know it learned the rule, not the answers.

  4. Play it

    Open Hunch's Signal Sort and clear Beeps and Boops.

Play the mission

This mission continues in the factory minigame.

Open the minigame →

Check your understanding

1. Why test on examples the model never saw during training?
2. You mislabel half the training examples. What usually happens?
Guest mode: your result is saved on this device only.