If an LLM could be "conscious in an LLM way", then why not the same, mutatis mutandis, for an ordinary computer program?

because an ordinary program is deterministic, LLM is probablistic + it has some synthetic-reasoning ability