Awesome World Model Hub Papers · Datasets · Projects
← Back to papers

Do multimodal models imagine electric sheep?

arXiv 2026 50.4 method

TLDR

Multimodal models develop mental imagery when solving spatial puzzles, forming an imperfect visual world model without explicit supervision.

Reasoning

The paper presents a novel finding that VLMs encode visual intermediate states as a byproduct of action prediction, supported by empirical gains from integrating visual tokens. However, the claim is limited to specific puzzle tasks and lacks comparison to explicit world model training.

Read-first score

Read-first score 50.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 43.

Recency 6%
100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%
71.7

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.71699013

Topical relevance 29%
61.4

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 18%
40

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=none

Reproducibility 18%
38

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code

Citation velocity 12%
0

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FrontierBridge

Rank sensitivity

Stability: volatile; rank range: 329.

Keyword Scores

world model
9
world dynamics prediction
8
interactive world model
7
world simulator
6
model-based reinforcement learning world model
6
generative world model
5
video world model
2

Deep Analysis

Innovations

  • Demonstrates that large multimodal models develop mental imagery as a byproduct of learning to select correct actions, without explicit visual supervision.
  • Proposes two methods to sharpen and utilize the mental images formed by the model.
  • Shows that integrating as few as sixteen visual tokens per step into the chain of thought improves solve rate from 83% to 89%.

Methodology

The authors fine-tune a Qwen3.5 VLM on twelve diverse visual reasoning tasks (tangram, jigsaw, sokoban, 3D mental rotation, rush hour, etc.) that require geometry, spatial relationships, and action consequences. The model is supervised to predict the open-loop sequence of actions from an initial state, and activations are analyzed to assess encoding of intermediate visual states.

Key Results

The model's activations after each action encode meaningful visual information about the intermediate state, indicating formation of an imperfect visual world model. Adding sixteen visual tokens per step into the chain of thought raises the average solve rate from 83% to 89%, with strong gains on reasoning-heavy tasks like jigsaw and 3D mental rotation.

Tags

multimodal modelsvisual reasoningspatial reasoningworld modelsmental imageryfine-tuningCVAI