Experts Have World Models. LLMs Have Word Models.

Summary

Latent Space published a guest essay by Ankit Maloo on February 7, 2026 arguing that the core limitation of current LLMs is not intelligence but "simulation depth" — the ability to accurately model how a response will land in a world full of agents with their own models, incentives, and triage heuristics. The essay distinguishes between "word models" (systems that evaluate text in isolation) and "world models" (systems that can simulate how text will be interpreted by recipients in context). The