Sins are just a metaphor on human (animal) impulsiveness due to faulty instinct; the Church just said 'think before you act, you might screw up greatly'. LLM's are like loaded dice in nature, thus, the output will never be reproducible but pretty much close enough in the 99% of caases, like a Markov based chatbot on stereoids (small LLM's and a GB trained Hailo will mainly converge in how to behave internally).
LLM based products are actually much closer to the Christian 'devil' -human prejudices causing bad acts, instead of slow analysis with reproducible proofs- than anything else. Because over time the accuracy gap without human revision will just get bigger and bigger upon feeding LLM's between themselves.