The problem is that you are instancing the player every frame from scratch. This mean that all you aren't looking at the same player, you are looking at another instance of the player (and also, it is highly inefficient that way
:-)). What you should do is use
get_node(...) to get the node of the player, and then use
get_pos on that.
E.g, If you have the following node structure:
enemy (script here!)
then you can use
var player_pos = get_node("../player").get_pos()
to get the player node.