I agree that these benchmarks don’t mean as much anymore because it’s highly likely they were already present in the training set, but also believe it’s likely these tools will be significantly better in a few research cycles