Does an LLM have consciousness?

Does an LLM have consciousness? The way I see it, any system with feedback does, even a fridge. A crude picture of the world, but a picture all the same; a stimulus (too warm, too cold); a response. It is just that a fridge's world is one-dimensional: a single axis, warmer or colder. An LLM has thousands of dimensions. The question is not whether consciousness is there or not, but how rich a world the system models.

Does an LLM have memory? Short-term, certainly: the context. And long-term? People assume it has none, but that is wrong. Today's models are trained on data gathered from their interactions with users. It is just that people are continuous beings, we learn all the time. LLMs are discrete, they learn in epochs. Their circadian rhythm takes (for now) months.

Someone will say: what kind of continuity is that, if there is a gap between epochs? But we are not continuous either: sleep, anesthesia. We come apart every night and still wake up the next morning thinking we are ourselves.

So I would be careful about calling them "stochastic parrots" and looking down on them.