The error message regarding the maximum number of arguments is wrong in broadcast:

>>> x = np.arange(3)
>>> args = [x] * 32
>>> np.broadcast(*args).shape
(3,)
>>> args = [x] * 33
>>> np.broadcast(*args).shape
Traceback (most recent call last):
...
ValueError: Need at least two and fewer than (32) array objects.

The point is that 32 still works, while it shouldn't according to the error message. Correct would be to require at most 32 array objects. Not sure why 32 is in brackets.