Error in Video 35. PyTorch Tensors and NumPy

[xiaolongli]

In the video, it is said that when converting between PyTorch tensor NumPy array, they don't share memory. And this is demonstrated by the following:

tensor = torch.ones(5)
array = tensor.numpy()
print(tensor, array)
tensor = tensor + 1
print(tensor, array)

This is a false proof as tensor = tensor + 1 doesn't modify tensor in place. In fact, the following snippet shows that the array does share memory with the tensor:

tensor = torch.ones(5)
array = tensor.numpy()
print(tensor, array)
tensor.add_(1)
print(tensor, array)