Yes, I frequently violate this myself, but at least I’ve been trying to keep it down.
If you do violate the rule, then make the new word as self-explanatory as possible. “Seed AI”, “Friendly AI”, and “neurohacking” are good. “External reference semantics” is bad.
Unfortunately following Eliezer’s advice seems to perhaps do the most to create the issues being considered about jargon here, because the more readily comprehensible jargon seems on first hearing it the more likely it is that it will be misremember and misapplied later (though “Schelling point” seems a notable case of something with no false-friend interpretation that gets misused anyway).
Eliezer also mentioned this in his old article on writing advice:
Unfortunately following Eliezer’s advice seems to perhaps do the most to create the issues being considered about jargon here, because the more readily comprehensible jargon seems on first hearing it the more likely it is that it will be misremember and misapplied later (though “Schelling point” seems a notable case of something with no false-friend interpretation that gets misused anyway).