Boltzmann-GPT: Bridging Energy-Based World Models and Language Generation

Abstract: LLMs generate fluent text, yet whether they truly understand the world or merely produce plausible language about it remains contested. We propose an architectural principle, the mouth is not the brain, that explicitly separates world models from LLMs. Our architecture comprises three components: a Deep Boltzmann Machine (DBM) that captures domain structure as an energy-based world model, an adapter that projects latent belief states into embedding space, and a frozen GPT-2 that provides linguistic competence without domain knowledge. We instantiate this framework in the consumer review domain using Amazon smartphone reviews. Experiments demonstrate that (1) conditioning through the world model yields significantly higher sentiment correlation, lower perplexity, and greater semantic similarity compared to prompt-based generation alone; (2) the DBM's energy function distinguishes coherent from incoherent market configurations, assigning higher energy to implausible brand-price combinations; and (3) interventions on specific attributes propagate causally to generated text with intervened outputs exhibiting distributions statistically consistent with naturally occurring samples sharing the target configuration. These findings suggest that even small-scale LLMs can achieve consistent, controllable generation when connected to an appropriate world model, providing empirical support for separating linguistic competence from world understanding.