If I were to create a new language I would get rid of the antiquated tradition of zero-counting (starting arrays at zero).
That would prevent annoying bug traps, like remembering that e.g. "sum(array, 1)" actually sums the array along the second dimension, not the first (yes, Python/NumPy really does that).
Zero-counting was fine for pointer arithmetic. Any language that doesn't use it should move back to 1-counting, as Matlab did.