[catach]

> There clearly is intelligence in these models

I would say instead that there's a level of complexity in these models that blow past our uncanny valley threshold heuristics for recognizing conscious intent. We've managed to find the edge of the usefulness of that hardwired sense.

I suspect the most challenging distinction for people is going to be: even if we do grant that a LLM had developed a kind of mind, our interactions with it are not actual communication with that mind. Human words are labels to organize sensory data and our experiences. The tokens an LLM works with have no meaning to it.