Play Ludo against a neural network
AlphaLudo is a 587,306-parameter graph transformer that learned Ludo by imitating the best model before it, then by playing against past champions and itself. It runs right here in your browser: no server, no account, no tracking.A 587,306-parameter network that learned Ludo, running right here in your browser. No server, no account.
01 Chance
Every turn starts with a roll you don’t control. What you do with it is the whole game, and it’s the only part a network can learn.
02 One move
It doesn’t look ahead. In a single pass it scores every square you could move from, all 225 of them, and plays the best. About 13ms, on your device.
03 How it sees
Earlier versions looked at the board like a picture. This one treats every square as a node carrying three numbers, and its attention is nudged along the paths the pieces actually travel.
Inside the model04 Your turn
It beats the model it replaced 52.7% of the time, and the strongest scripted player 68.4%. Against you, the dice still have a say.
It knows its odds
Besides choosing moves, the network estimates who will win. This is one real game of AlphaLudo against itself: it rated green at 78% at best and 9% at worst, and the dice swung it more than once. You see the same estimate live while you play.
Your turn
Green is yours. AlphaLudo plays amber, and after every move it tells you how it rates your chances.
Take your seatThe board is empty.
Loading the model…