Or, given https://thegradient.pub/othello/, why wouldn't it have an internal model of chess? It probably saw more than enough example games and quite a few chess books during training.