Also, note that the strong version of this argument mostly applies to hypothetical future short-timelines-world RSI'd LLMs of mid-2027 or thereabouts, not to modern ones.
I don't think this would interfere with a basic demonstration.