[unknownx113]

Using predictions of past events as the benchmark where the LLMs have already been trained on the results seems quite flawed to me