"Pro: LLMs have passed the Turing test. If you have a black box containing either a human or an LLM, and you interrogate it about consciousness, it's quite hard to tell which one you're talking to. If we take a human's explanation of their own conscious experience as important evidence of consciousness, then we must do the same for an LLM."
You mean unless you run its output through Pangram?
The thing about LLMs is how reducible they aare to deterministic processes. If you had the input text, exact model architecture, every weight & bias, the tokenizer, and the precise arithmetic rules of the implementation, plus a random-number sequence, then you could perform every operation with pen and paper and eventually get the same output as if you had typed the sucker into the input box on Claude. It would just take an insanely long time. If you gave a few billion clerks ginormous books containing the weights and taught them fixed-point arithmetic, and set up a mail conveyance bureaucracy corresponding to the transformer layers, and gave them enough paper, you might be able to get one output in a couple hundred years.
The fact that you could in theory run a LLM by hand, though, suggests to me that these things aren't conscious.
"Pro: LLMs have passed the Turing test. If you have a black box containing either a human or an LLM, and you interrogate it about consciousness, it's quite hard to tell which one you're talking to. If we take a human's explanation of their own conscious experience as important evidence of consciousness, then we must do the same for an LLM."
You mean unless you run its output through Pangram?
The thing about LLMs is how reducible they aare to deterministic processes. If you had the input text, exact model architecture, every weight & bias, the tokenizer, and the precise arithmetic rules of the implementation, plus a random-number sequence, then you could perform every operation with pen and paper and eventually get the same output as if you had typed the sucker into the input box on Claude. It would just take an insanely long time. If you gave a few billion clerks ginormous books containing the weights and taught them fixed-point arithmetic, and set up a mail conveyance bureaucracy corresponding to the transformer layers, and gave them enough paper, you might be able to get one output in a couple hundred years.
The fact that you could in theory run a LLM by hand, though, suggests to me that these things aren't conscious.