Even then…

When AI preaches you the Gospel

When getting ready for a meeting of our congregation, I was reviewing the articles we were planning to read at that meeting. Usually, everybody is encouraged to provide brief commentaries about the material we study during the meetings. In fact, the articles themselves have questions for every paragraph, and the person leading the study asks these questions and actively solicits responses and commentaries.

So I decided to use AI/LLM to generate responses to these questions. This is really what generative LLM is good at writing a coherent text based on the information it has learned during the pre-training and the information in the context. Plus, the training data on which the model was pre-trained for sure included the Bible, and also likely other religious books.

I started with a simple prompt, and then subsequently refined it several times. It seems that the responses are either very concise or chatty. I think the biggest problem I saw was that the generated text did not sound like me. I mean, people know what I am likely to say. Having a super-polished text that was not written by me would raise some questions. Since I was running out of time, I took the text and rephrased it myself.

The commentary was very well received during the meeting. I did not disclose its source because it does not matter. On the one hand, I did work on it. On the other hand, we do not look for any original content, because everything we say must be based on Scripture.

As a side note, Paul said in 1 Corinthians 14:21: “‘With the tongues of foreigners and with the lips of strangers I will speak to this people, and even then they will refuse to listen to me,’ says Jehovah.” Can it be now changed to: “‘With AI-generated text I will speak to this people, and even then they will refuse to listen to me,’ says Jehovah.”