The program - yes, it is a rule-based program. But the reasoning and logic responses are not implemented explicitly as code, they are supported by the network and encoded in the weights of the model.
That's why I see it as not bounded by computability: LLM is not a logic program finding perfect solution to a problem, it's a statistical model to find next possible word.
That's why I see it as not bounded by computability: LLM is not a logic program finding perfect solution to a problem, it's a statistical model to find next possible word.