This doesn't follow, at least not in my understanding. Consider the following: Qualia are "what it 'feels like' to experience some sensory input." Up until r

This doesn't follow, at least not in my understanding. Consider the following: Qualia are "what it 'feels like' to experience some sensory input." Up until recently, most LLMs were "once through" meaning that the only "sensory inputs" they "experienced" would be the raw text. So we might argue that "experiencing sensory input" means "tokenizing raw text," and that therefore the tokens that the LLM processes internally are the qualia. But that's un-satisfying. We don't say that the impulses sent from the eye to the brain are the qualia, and the tokenization process sounds more like "eye turning light into electrical signals" than what we actually mean by qualia. So now we focus on the "feeling" word in our definition of qualia. A feeling isn't a token or an electrical impulse, its our internal reaction to that token or electical impulse. So because once-through LLMs have no input that corresponds to "their internal reaction to a token", they can never be said to "experience" a "feeling" using our previous definition of experience as "processing some input". But this directly suggests the solution to the qualia problem: if we were to build an LLM that did accept an input that represented "its internal reaction to the tokens it previously experienced" then we'd have invented qualia from scratch. The qualia would be precisely the log file that the LLM generated and "sent back around" as input for the next round.