I think that the type of people who can easily pick up subtext have come to rely on that channel of communication and don't realize they need to be more direct and verbose when chatting with language models.