Nice article. But it's important to recognize that ChatGPT does not aggregate and weight sources. Instead, it's "just" a probabilistic sentence completer. It appears to be neutral because (1) these sentences are learned from diverse sources, (2) neutral introductory language is more frequent in the training dataset, (3) the human feedback prioritized pleasantness, (4) within a session, it maintains continuity. In reality, the point of view you will get from chatgpt is pretty random.