Knockout: computed observable vs function

When using knockout, what is the advantage of using read-only computed observables rather than simple functions?

Take the following viewmodel constructor and html snippet, for example: ​

var ViewModel = function(){
    var self = this;
    self.someProperty = ko.observable("abc");
    self.anotherProperty = ko.observable("xyz");
    self.someComputedProperty = function(){
        return self.someProperty() + self.anotherProperty();
    };    
};

<input data-bind="value: someProperty"/>
<input data-bind="value: anotherProperty"/>
<p data-bind="text: someComputedProperty()"></p>

Everything here seems to work as you'd expect, so is there a reason why I should instead use:

​var ViewModel = function(){
    var self = this;
    self.someProperty = ko.observable("abc");
    self.anotherProperty = ko.observable("xyz");
    self.someComputedProperty = ko.computed(function(){
        return self.someProperty() + self.anotherProperty();
    });    
};


<input data-bind="value: someProperty"/>
<input data-bind="value: anotherProperty"/>
<p data-bind="text: someComputedProperty"></p>

I notice that the documentation at http://knockoutjs.com/documentation/computedObservables.html states that "...declarative bindings are simply implemented as computed observables", so does that mean there's need for me to use them explicitly in my viewmodels?