I think that’s the biggest thing that I take from this: any text should at least hint at the rich tapestry of things it is resulting from, if not directly discuss it or link to it. A tapestry not just made from other texts, but other actions taken (things created, data collected, tools made or adapted), and people (whose thoughts you build on, whose behaviour you observe and adopt, who you interact with outside of the given text). Whether it’s been GPT-3 generated or not, that holds.
Useful and likely human written texts show the richness of the context it results from, by showing and linking. Not just to/with 1) other texts, but also 2) other actions (things created, data gathering, experiments, tools adapted) and 3) people (that provided input, you look at, interact with outside the text). Even if such things were generated following up those leads should show its inauthenticity.