This interpretation would have been ok for old generation models without search tools enabled and without reliable tool use and reasoning. Modern LLMs can actually look up the existence of papers with web search, and with reasoning, one can definitely get reasonable results by requiring the model to double check that everything actually exists.