Experts Have World Models. LLMs Have Word Models

While large language models can produce professional-looking documents and messages, they fundamentally lack the adversarial reasoning skills that domain experts possess, operating more like "word models" than true world models. Unlike experts who automatically simulate how opponents will react, exploit weaknesses, and adapt strategies in multi-agent environments, LLMs evaluate text in isolation rather than considering the strategic dynamics where other parties actively model and counter their moves.