I'm disappointed he is open to Anthropomorphism, I had led myself to believe he wasn't that minded. He also appears (from this article) to think the predictive quality of text models implies knowledge of the world, arguing that since we are also trapped inside a sense factory filter, we can't exclude that what the LLM does is also bound by its sensory inputs.
I just don't agree. This is Searle's chinese room brought into the modern computing era. Inferences about how the system behaves don't actually lend themselves well to discussions about where cognition lies in the system, if at all. Informatiom is encoded, sure. And inferences are shown to be drawn based on the information model. The inferences are to my mind still shallow, for indicating intelligence.
Still, good on him for being lucid, and for deciding to quit google to maintain his independence of thought and action stemming from it.
I just don't agree. This is Searle's chinese room brought into the modern computing era. Inferences about how the system behaves don't actually lend themselves well to discussions about where cognition lies in the system, if at all. Informatiom is encoded, sure. And inferences are shown to be drawn based on the information model. The inferences are to my mind still shallow, for indicating intelligence.
Still, good on him for being lucid, and for deciding to quit google to maintain his independence of thought and action stemming from it.