Did you read the article or just the title? They mention the specific models the researchers were testing and note that increasing model size did not seem to offer much improvement on this metric. It also ends with a discussion of research into methods for improving performance on queries involving negation.