arxiv:2507.05169

Critiques of World Models

Published on Jul 7

· Submitted by

Authors:

Abstract

World Model, the supposed algorithmic surrogate of the real-world environment which biological agents experience with and act upon, has been an emerging topic in recent years because of the rising needs to develop virtual agents with artificial (general) intelligence. There has been much debate on what a world model really is, how to build it, how to use it, and how to evaluate it. In this essay, starting from the imagination in the famed Sci-Fi classic Dune, and drawing inspiration from the concept of "hypothetical thinking" in psychology literature, we offer critiques of several schools of thoughts on world modeling, and argue the primary goal of a world model to be simulating all actionable possibilities of the real world for purposeful reasoning and acting. Building on the critiques, we propose a new architecture for a general-purpose world model, based on hierarchical, multi-level, and mixed continuous/discrete representations, and a generative and self-supervision learning framework, with an outlook of a Physical, Agentic, and Nested (PAN) AGI system enabled by such a model.

View arXiv page View PDF Add to collection

Community

mingkaid

Paper submitter about 15 hours ago

A world model (WM) is NOT about generating videos, but IS about simulating all possibilities of the world to serve as a sandbox for general-purpose reasoning via thought-experiments. This paper proposes an architecture toward that.

In this paper, we formally show that WM is an integral part of an optimal, general agent. We then review several schools of work on WM toward this goal, and propose our alternative, the PAN (Physical, Agentic, Nested) architecture for general WMs.

PAN is built on the following principles:

use data from all modalities of experience;
employ a mixed continuous and discrete representation;
adopt a hierarchical generative model paradigm with an extended-LLM backbone (for discrete concept-based reasoning), as well as a generative embedding predictive module (for continuous gradient-based reasoning);
train over a generative loss grounded in observation data; and
apply WM to simulate experiences for training agents using reinforcement learning.

We will soon release a 27B 1st-version of PAN, and it will be the first playable general-purpose world simulator. Stay tuned!