Isn't this the claim for which they link to a paper [0] directly, which describes an algorithm by Ryan Williams for achieving this bound?

[0] https://www.sciencedirect.com/science/article/pii/S030439750...