Considering how weirdly detuned tokens selections have become in Anthropic's LLM prose in recent models, there is a chance this goes unnoticed in everyday use.