Grok (formerly Numenta) has a slightly more technical white paper that goes into more detail on the actual algorithms from On Intelligence:

https://www.groksolutions.com/htm-overview/education/HTM_Cor...

There are some subtle differences between HTMs and straight up deep learning, mainly the requirement for HTM data to be temporal and spatial.

I know Andrew used to sit on an advisory committee at Numenta, I don't know if he still does.