Yeah that video was 8 months ago but they were using their own testing pilots as you can see in the video. The first part is basic crash testing and object load/streaming stress.
They started in on the real stuff in June after the MS Co-Pilot announcement from March. That last part where Todd is talking about a scripted bot that can run through the beginning quests - that has expanded through machine learning and they have over 2,000 instances of the game being played in all aspects gauging scripting errors, typos, object clipping, Z-Axis alignment, memory leaks, streaming issues, core dumps, frame testing, frame limiting, tears - to name a few.
Will be interesting to find out here in about 3 days for early access how it all panned out.