My dad was talking to me about his new best friend ‘Chat’, as he calls it. He loves #ChatGPT because he says he has the deepest conversations he can remember having with it about religion, a passion of his. In particular, he says it can speak at a scholar’s level about the Bahá’í faith, our families religion. At first, I was a bit surprised because most of the benchmarks I am familiar with test performance on math, science, coding, and medicine. I hadn’t heard from anyone that ChatGPT was particularly deeply knowledgeable about theology.
But then I thought, well most religious text is openly licensed and repeated all over the internet, so I suppose quite a lot of ChatGPT training data must come from published religious texts? There must be tons of it in Project Gutenberg, OpenCrawl and the Internet Archive. I can’t find much on this other than some research into its religious bias.1