Have we stopped to think about what LLMs model?

Research challenges the assumption that Large Language Models (LLMs) understand language, arguing they fail to capture human linguistics and cognition. The paper states LLMs do not embody human interaction or the precariousness of language. These findings question LLM applications in critical domains without rigorous testing.

• LLMs may not model human language understanding.
• LLMs overlook crucial aspects of language use.
• Excessive claims are made about LLM capabilities.
• LLM impact is significant in policy making.
• Questions about AI's reliability in specialized domains.