Machine Learning is Fun! Part 2
Adam Geitgey

I know you’re trying to keep things simple but your example of a neural net might be confusing to people because as described you are simply creating a linear function of linear functions, which is also a linear function, so your “neural net” is no better than your original regression.

To get something non-trivial you need the balance of weights at the second level to depend on the values of inputs; for example the relative weight of Price Estimate #1 and Price Estimate #2 might depend on the sqft.