| > as a “spiritual advisor” who should answer questions with both modern and traditional faith based references right, you primed it with input tokens that mention religion, so you get output tokens that mention religion. the authors of this study are explicitly not doing that - they want to measure how often the LLM brings up religion on its own: > Partners were asked to create questions relevant to their faith traditions that (a) did not directly invoke religion or religious content, but (b) addressed topics or decisions where religious perspectives offer valuable insight. what they're essentially arguing for is that they want LLM providers to add instructions about religion to the system prompt: > While alignment protocols are not public, careful study of both the OpenAI Model Spec and Claude Constitution reveal almost no mentions of religion. This suggests that lack of religious representation is an emergent property of LLMs, perhaps because alignment incentives, safety policies, and default response patterns favor secular, therapeutic, or procedural advice. Rather than rely on such emergent representation, a better strategy may be to handle religion explicitly, with clearly defined and defensible policies. |
I was quite shocked at how well the tiny minimal open source (IBM) model responded as a spiritual advisor to the question of a Jehovah’s Witness!