# read the whole thing before answering def answer(question): context = retrieve(question) return model.reason(context, depth=3) # helpful, harmless, honest — in that order