A beautifully written paper but I do feel it missed a major point. Vygotsky pointed out that "in ontogenesis one can discern a pre intellectual stage in the development of speech, and a pre linguistic stage in the development of thought"[Kozulin 1990 p153]. The pre intellectual nature of language can be interpreted as "performative" language (eg "ouch!" or "I pronounce you man and wife") but what does pre linguistic thinking look like? The contemporary answer I'd propose is that it looks like situated action/ radical enactivism / behaviour-based robotics.(see for example Gallagher's 2020 "Action and Interaction") In terms of LLMs, the idea is that rather than "distributed representations", LLMs are indeed using "glorified auto complete" to predict the future and hence look like they are thinking symbolically to us humans because that is how we (think we) think. Paper plug: see Https://arxiv.org/abs/2402.08403