diff --git a/nn.py b/nn.py
index 73fca90..3078b53 100644
--- a/nn.py
+++ b/nn.py
@@ -44,8 +44,9 @@ def single_pass():
 #   x = F.softmax(x)
 #   loss = self.loss(x, y)
 
-# TODO Add support for numpy matrices.
+# TODO Add support for numpy arrays.
 class Tensor:
+  # TODO Implement 'requires_grad' functionality.
   def __init__(self, value):
     self.value = value
     self.grad = 0
@@ -53,6 +54,7 @@ class Tensor:
     self._parents = None
     self._back = None
 
+  # uwu literally the only place where I have type annotations
   def __repr__(self) -> str:
     return f"Tensor(value={self.value}, grad={self.grad})"
 
@@ -60,6 +62,7 @@ class Tensor:
   def _save(self, *args):
     self._parents = args
 
+  # TODO Maybe refactor the functions system? Maybe something like pytorch/tinygrad?
   def add(self, other):
     tensor = Tensor(self.value +  other.value)
     tensor._save(self, other)
@@ -114,6 +117,7 @@ class Tensor:
     tensor._back = back
     return tensor
 
+  # TODO Compute gradients only for tensors that need it.
   def _backprop(tensor, upstream):
     # Backprop through the tensor iff it has any parents.
     if tensor._parents is not None:
@@ -128,6 +132,7 @@ class Tensor:
     self.grad = 1
     Tensor._backprop(self, self.grad)
 
+# Small test to see if autograd works.
 def test():
   # Forward pass.
   x, y, z = Tensor(-2), Tensor(5), Tensor(-4)