Tag
The author criticizes the common practice of including cached input tokens in discussions about token usage in AI models.