AI Influence Spreads: ChatGPT Now Alters Spoken Language, Not Just Text

Researchers at the Max Planck Institute have discovered a surprising trend: Chatgpt not only changes the way we write, but also how we speak. The analysis of over a million podcasts and videos clearly showed this.

How AI shapes our language

Since its introduction at the end of 2022, Chatgpt has developed into the fastest growing consumer product in history and has a significant influence on how we learn, write and work. After only two months, the AI application reached 100 million active users – a record for which Instagram needed two years and Netflix over ten years. Now a new study shows that the AI also changes our (written but also spoken) language in an unexpected way.

A team of researchers from the Max Planck Institute for Educational Research initially analyzed millions of emails, essays and scientific work that were created with chatt. They identified certain words that the AI system prefers-so-called “GPT words”. These characteristic terms are characterized by their formal, often somewhat artificial expression, which is typical of large voice models.

Significant changes in language use

The researchers then examined over 360,000 YouTube videos and 771,000 podcast episodes for the frequency of these (English-language) terms. Like the scientists in theirs arxive published study (PDF) Explained, there were remarkable trends: The word “delve” (fathoms) proved to be particularly meaningful linguistic indicator. Its use rose significantly in unwritten podcast talks about science, business and education – although these were spontaneous, non -prepared conversations.

Other typical GPT words such as “understand),” Swift “(fast),” Meticulous “(meticulously) and” intricate “(complicated) (complicated) listed in academic lectures and discussion rounds of 25 to 50 percent per year. The increase in terms such as “Showcase” (presenting) and “Underscore” (underline), which chatt often used in his answers, was particularly striking. Unlike earlier technologies (which always have an impact on people), KIS work bidirectional: they learn from human language and influence them at the same time.

Standardization risks

The scientists warn of a gradual standardization of tone and style. While Chatgpt was trained to communicate politely, neutral and structured, human language lives from their peculiarities and, above all, imperfections: regional peculiarities, hesitation, emotional outbreaks and spontaneous wording. These characteristics make the language alive and authentic – and this threatens LLMs a lot.