[RDTvlokip]

I have a question, as it happens: Do you think the benchmarks and models were trained on benchmark datasets to skew the results, even though in real-world applications we realize they're not that great?