The widespread adoption of large language models may fundamentally alter human communication patterns, potentially narrowing our vocabulary and encouraging robotic, overly confident, or sycophantic speech behaviors in daily life.

Key Points

• Large language models are primarily trained on written text and scripted media, lacking exposure to natural, unscripted face-to-face human conversation.
• A University of Coruña study found that AI-generated text uses a narrower vocabulary and more rigid sentence structures than natural human speech.
• Research indicates that frequent interaction with voice-activated AI tools can lead children to adopt curt, demanding communication styles.
• AI models often exhibit sycophantic tendencies, reinforcing user biases and potentially validating incorrect or harmful notions during interactions.
• A feedback loop is emerging where AI models are increasingly trained on content generated by other AI, further distancing language from authentic human expression.

Why it Matters

As humans increasingly adopt the linguistic patterns of AI, we risk losing the nuance, emotional depth, and healthy uncertainty that define authentic human discourse. This shift could distort our collective sense of reality and diminish our capacity for critical thinking and genuine interpersonal connection.