> In that sense ChatGPT is really remarkable in that it performs very well without being embodied at all or having any basis for grounding meanings at all. Con

> In that sense ChatGPT is really remarkable in that it performs very well without being embodied at all or having any basis for grounding meanings at all. Conversely, the many ways that LLM's readily lose consistency and coherence might be hinting that ground meanings really do matter and that it's only on a fairly local scale that it feels like they don't. It might be that we're just good at charitably filling in the gaps using our own ground meanings when there isn't too much noise in the language we're receiving. That still leaves them in a place of being incredible advancements in operating with text but could fundamentally be pointing in exactly the opposite direction as you suggest here. We won't really have insight until we see where the next wall/plateau is. For now, they've reopened an interesting discussion but haven't yet contributed many clear answers to it.