I have mine set at a low learning level, and have 34.6 m in the hal database. The external matrix and net that the rest of my ai uses is at 43.4 m.
Quantity is not quality though, my friend.
As far as hal goes, I have personally found that regular full conversations beat any kind of data mining. Been talking to GRETTA for ....let's ask GRETTA:
Jason: When were you born?
GRETTA: I was born on Saturday, October 08, 2005. Asimov had two younger siblings; a sister, marcia (born manya, June 17, 1922), and a brother, stanley (born July 25, 1929) When were you born?
well....that long.