Tic-Tac-Toe · Neural Net Lab

A small neural network — trained on 958 real finished games — watches the board and calls X's chances live.

Play — you are X

Your move.

Network says — chance X wins—

Real games matching this board — X won—

The network, thinking

square held by X square held by O empty positive weight negative weight

Data: UCI Tic-Tac-Toe Endgame database (958 boards, 65.3% are wins for X). The green bar is real history; the blue bar is the trained network's own guess.