This is even worse advice! If you have a small input, you want a hashing function that matches that size[1] of the input as closely as possible. Collisions aren't important here (so "strength" isn't important), just the randomness. MD5 and SHA families give you (mostly) random distributions. The bias in the results are practically meaningless.

[1] https://crypto.stackexchange.com/questions/12822/are-the-sha...