Study shows humans are adopting ChatGPT’s language in everyday speech

If you are what you eat, then maybe you’re starting to sound like what you prompt?

This is the first empirical evidence that spoken language is shifting in response to large language models. Researchers saw a clear spike in the use of ChatGPT-associated words following the chatbot’s public release in November 2022.

Statistical modeling showed increases of up to 51% in the use of words like “adept” and “delve” over an 18-month period.

And this is no coincidence, either. Researchers tested alternative change points (like one or two years before ChatGPT's release) and found no similar trends, ruling out broader cultural shifts as the cause.

The researchers have analyzed over 280,000 academic YouTube videos from 20,000 channels. Although the scripts weren’t AI-generated, the humans were sounding like they were reading from such.

They also found a strong correlation between how uniquely a word appears in ChatGPT-edited abstracts and how quickly it entered academic speech. In other words, the more a word sounds like ChatGPT, the faster people start saying it.

Experts are now questioning the long-term effects of this feedback loop: as we train artificial intelligence (AI) on our language, then mimic its outputs, the line between human and machine expression could blur – flattening the unique voices that give language its richness.

There’s also a concern about manners – as we grow used to being brief or blunt when writing prompts for AI, it could affect how polite and warm we are with other people.

The irony is hard to ignore – humans designed AI to sound more like us, and now, it seems, we're sounding more like it.