Something seems a little bit off in this cost-benefit analysis to me. You seem to compare the tiny tiny cost of delaying one breath to the sizable accumulated  impact of 1 billion people doing this for a year.  But that is not really helpful to get an intuition. The tiny tiny cost of delaying breathing once will also accumulate if 1 billion people do this for a year.

Of course, it is still possible that the accumulated cost is lower than the accumulated benefit. But in a way, this whole accumulation does not matter. All that matters is if the cost is higher than the benefit. 

Nice post!

Do you think a person working on this should also have some basic knowledge of ML? Or might it be better to NOT have that, to have a more "pure" outsider view on the behaviour of the models?

I personally think the risks of these videos are relatively low because they do not mention EA. People who are convinced by the ideas in the jokes might start a google search and eventually find EA. Those that feel disgusted by the jokes might just think "what an idiot" and stop there. I doubt they would go on to search about what this is all about, find EA, and then try to act against that.

Just wanted to let you know that this was super amusing to read (including the hyper-linked content)! Some nostalgia for time in high-school when I was translating this stuff in Ancient Greek class :D