Researchers found that LLMs often extrapolate from prominent historical data but struggle with more obscure details. For instance, GPT-4 incorrectly stated that scale armor was present in ancient ...