Standard datasets can no longer be used for benchmarking against LLMs since they have already been fed into it and are thus too well-known to compare to lesser known documents.