It looks like there is quite a bit of discrepancy in the cost values after performing gradient descent with some of the implementations done so far. In terms of credit, I'll give full credit to those who get the right Thetas and followed the directions. We will examine our code together tomorrow to get to the bottom of this. 

My own code had a bug. Here's my corrected values:

Cost before gradient descent:  305.236913265

Thetas:  [[ 23.44490618]
 [ -6.48710167]]
Cost after gradient descent:  9.33830885019