LLMorphism: When Humans Start Behaving Like Language Models

Remember the old joke about a programmer who starts talking to his wife in JSON? Well, it's not a joke anymore — it's a scientific term: LLMorphism. Researchers from several universities have noticed that people who actively chat with large language models gradually adopt their speech patterns: more formal constructions, avoidance of ambiguity, and even characteristic phrases like "as a language model, I cannot...".

The gist is simple: we adapt to our conversation partner, even if it's a neural network. It's like picking up an accent after a week abroad, only the accent is impeccable grammar and zero sarcasm. The study authors ran a series of experiments and confirmed: after a long dialogue with an LLM, subjects started using longer, more structured sentences, less slang, and even copied typical AI disclaimers.

On one hand, it's funny — we're learning to be more "correct" from machines. On the other, there's food for thought. If developers start writing code like ChatGPT, commits will be perfectly documented, but those "magic" one-liners that work for no apparent reason will vanish. And imagine JIRA tickets phrased as "As an AI assistant, I cannot estimate the time required" — that's a whole new level of bureaucracy.

But don't panic just yet. So far, LLMorphism only appears in those who chat with models for hours on end. If you just use Copilot for autocomplete, your coding style is safe. Still, check your latest Slack messages — has "According to my training data..." slipped in?

METABYTE studio's take: We're all for learning from technology, but if your product manager starts replying with "As an LLM, I cannot prioritize tasks without additional context" — maybe it's time to update not just the model, but also the company culture.