play the seven pillars of LOOMUS Distill
Seven tiny games. Each one is a playable demo of one principle from LOOMUS Distill — Metcalfe’s Law, Bayesian belief, multi-armed bandits, RLHF, Popper’s falsifiability, Kaizen compounding, Constitutional AI. Play the idea, then read the theory.
Goal: reach value = 100. Value = N² × how connected the network is. Isolated nodes = 0 value. You need to both add nodes AND connect them. Click empty space to drop a node. Click two nodes in a row to connect them.
Your goal: find the most generous mushroom in as few kicks as possible. Each kick gives a noisy hint about that mushroom’s true average. When you’re sure, hit “commit” — we’ll see if you were right. Stop too early → wrong guess. Kick forever → wasted time.
Plant one flower per day. That’s the whole game. Come back tomorrow, plant one more. Day 1 = a seed. Day 3 = a sprout. Day 7+ = a tiny bloom. After 30 days — a garden. Kaizen: small steps, no stopping — time does the work.
One box hides the prize. You can ask for hints — but each hint only shifts the probability, never confirms. After 3 hints, you guess. Knowing isn’t binary — it’s a distribution that sharpens with evidence.
Six claims. Sort each into KNOWLEDGE (can be proven wrong) or BELIEF (cannot be proven wrong). Karl Popper: if a claim can’t be falsified, it’s not knowledge — it’s a feeling wearing the costume of knowledge.
Goal: train the frog to match a target personality. The frog says random things in 4 styles. Like it? thumbs up → that style’s weight goes up. Don’t? thumbs down. After 10 rounds, your frog’s vibe distribution should match the target. This is RLHF in miniature.
Every AI assistant is shaped by hidden rules — its “constitution.” The cat below has 3 secret rules from the list. Her answer is the product of those rules.
Your job: read her answer, look at the 6 possible rules, and check the 3 you think she used. The list shows what each rule does to her output.
↓ which 3 rules made her say that? (check exactly 3)