I really like his idea of showing the dynamic range increase—good demo. However, he’s totally wrong about noise shaping, that’s not what it does. The term “noise shaping” was a bad term to go with from the beginning, because since dithering is adding noise, noise shaping sounds like it adds shaped noise. It doesn’t, it adds the same noise (usually—though you can skip that step for a different result), and the quantization error subsequently gets shaped. In math circles, “error” and “noise” are often synonymous (particularly when the error isn’t simply a bias).
My take on dither (for the brain):
And musically (practical considerations, value):
And yes, this has slipped off topic because the pink noise mixing technique is unrelated to dither. Basically, the technique limits the loudest frequencies of a track to the 3 dB/Oct slope of pink noise. Not going to win you any awards, but won’t sound awful given decent source tracks.