I think his point is that for some checksums it could be trivial (and for some, tools already exist). Checksums aren't designed for this, while on the other hand secure hashing is. As a result, authors of hashing algorithms often attempt to mathematically prove their strength and resistance to a collision.