Fork me on GitHub

A dice game. A live AI. Your move.

Crack the strategy.
Beat the machine.

Bank It turns a push-your-luck dice game into a live battle of probability and AI. Make your call, see what the machine chooses, and uncover the math that can give you the edge.

The rules take 30 seconds. Mastering the decision takes longer.

1 in 6 +184?
Round 8 / 10 AI connected
You620
VS
The Analyst590
Shared pot 184 points at risk
Bust chance1 in 6
Your lead+30
Rounds left2
AI decisionReading the table…

Quick—what would you do?

The pot is 184.
One roll could change everything.

Live scenarioRound 8 / 10
You620
+
AI590
Pot184points
Next-roll bust chance 1 in 6

Your move

There is no trick answer. Context changes the call.

1/6+18410R

Math isn’t the homework here.
It’s how you win.

Probability reads the risk. Strategy uses it. AI gives you someone to test it against.

Easy to start. Hard to solve.

Three moves.
One strategy problem.

The dice create the problem. You decide how to solve it.

01

Roll

The shared pot grows with every safe roll. Doubles can send it soaring.

More reward means more to lose.
02

Read the risk

Watch the pot, the score, and the rounds left. Context changes the call.

Probability becomes strategy.
03

Bank or push

Lock in the pot—or risk a seven wiping out everyone who has not banked.

No perfect move without context.

Question the AI

Same table. Same facts.
Different calls.

Every opponent sees the same pot, scores, and rolls. Different models can still weigh the risk differently. That does not make one automatically right.

Shared game state
Pot211
PositionTrailing by 85
Rounds left1
Bust chance1 in 6
Legacy generationGPT-3.5
DecisionBank
Risk profileUnpredictable

Locks in what it has, even though 211 points may not be enough to close the gap.

  • Sees the large pot
  • Weights immediate loss heavily
  • Underreacts to the final round
Human-built policyState Delta
DecisionRoll again
Risk profileAdaptive

Raises its target when trailing and refuses to bank a score that cannot take the lead.

  • Measures the score delta
  • Adjusts for time pressure
  • Uses tested thresholds
See why the models disagree

Go behind the game

Three ways to crack it.

You do not need to memorize a formula. Start with a move you want to understand.

01
Probability
16chance of a bust

Read the odds

Find out what “one in six” really means when 184 points are on the line.

Run the same roll 36 times
02
Artificial intelligence
BANKROLLROLL

Question the machine

Compare model calls and find the situations where stronger AI still makes weak moves.

Compare three AI calls
03
Simulation + code
10,000games tested

Test the strategy

One lucky game proves nothing. Run thousands and see which ideas keep winning.

Simulate 10,000 games

Luck gets one game. Strategy gets 200,000.

We tested the machines.

The strategy work behind Bank It compares AI-generated formulas and human-built policies across hundreds of thousands of simulated games.

Don’t trust the smartest-sounding answer. Test what actually works.
See the experiment
200K game showdownTie-adjusted wins
01State Delta
53.45%
02GPT-5 strategy
31.83%
03GPT-4o strategy
12.88%
04GPT-3.5 strategy
1.84%

One reproducible 200,000-game, four-strategy simulation. Results do not describe every possible matchup.