One of my goals in life before dementia sets in for real is to devise some model, perhaps a conceptual framework which will allow us to escape the clutches of habitual simplification, a subset of which is dichotomous thinking, which in turn leads to the inevitable painting of strawmen as a way to prove our point (among other things). How sweet it would be to shortcut all the mandatory twists and turns of discourse that follows: "this is a mischaracterization of x", "not all x are y", "x and z are really not opposites, but overlapping", "this is a spectrum with a bell curve, not an either/or" etc. But of course, we all do this, not because we can't think clearly, but because we have an agenda, or maybe more frequently: want to trash talk the stance of an opponent because of what the proliferation of that stance might lead to, and so on. Taking such into account should be an integral part of the conceptual framework, obviously.
In the case at hand, one could easily argue that people in the debate are creating false dichotomies: LLMs are either stochastic parrots OR algorithms with an understanding, when in reality they are both (and also something else completely), but acknowledging such would likely require that one doesn't have an axe to grind, a stake in the field or what you might call it. It would require extending some "philosophers charity" to an opponent, that maybe has tried to undercut one's work for decades, in a field steeped in fierce and bitter competition for a name, like academia. Or, in case one has a business in the field, it would require maybe saying something that puts your core business idea in the crosshairs of legislators, or something else that doesn't serve your long term business interests.
Which brings us to this important aspect of this "conceptual framework against simplification" already briefly touched upon, namely identifying the bias of the participants in the debate. My impression is that naming bias has largely gone out of fashion, which is a pity because it is really a necessary part of understanding an argument: it rarely explains it all (that would be a grave simplification), but it is really a vital part of understanding an argument. And a difficult one, because people will go to extreme lengths to hide their agenda. And the current conceptual framework for unravelling bias has largely been occupied by the fact-checking industry: i.e. things are either true or false, and once you are cleared (like most mainstream media) then bias is not questioned. But we can be assured, there is always some bias, and it is usually relevant to name it (if one can see it), even if it infuriates the named party.
Just sayin'.