In the abstract of the paper you quote, they say (edited by me a bit to make it more brief:
First we show that a neuron with several thousand synapses...can recognize hundreds of independent patterns of cellular activity... We then propose a neuron model where patterns detected on proximal dendrites lead to action potentials...and patterns detected on basal and apical dendrites act as predictions... We then present a network model based on neurons with these properties that learns time-based sequences.
Basically, they are showing a bunch of things that are true in some simulations they have run, which are based on some biologically plausible assumptions. There is no reason to doubt their simulations, and Hawkins is a credible person.
The basic assumptions they make are grounded in known and agreed upon concepts in neuroscience: the anatomy of different cell types, for example, the effects of synapses at different places in the cell. The whole synthesis of their work, though, is just a hypothesis, and that's also how they are presenting it. It wouldn't make sense, and Hawkins would not claim, that this is "accepted": it's a proposed idea that they show some properties of in silico. Even if some cells in the brain do work this way, it doesn't necessarily mean it is an organizing principle, either. If you're interested, Hawkins' book On Memory tells a bit more.
You can add a bit of skepticism if you'd like because these authors work for a company that sells patented machine learning technologies based on their neuroscientific work, but I don't think that's a reason to throw them out entirely, and their academic publications seem sound. I like to use Google Scholar to track citations of a given paper; if you follow that link or find the paper there on your own, you'll see it has been cited ~150 times, though many of those citations are coming from a more computational/machine learning angle, which makes sense: that's the main application for the paper. It's not really proposing any novel biology, since the biology is mostly based on things already known.