The author argues that different AI models have varying success rates in self-detecting their own content, with Claude being an outlier due to its difficulty in self-identifying generated text.