MetaOthello: A Controlled Study of Multiple World Models in Transformers

arXiv 26.2 2026 36.1 method, benchmark

TLDR

Transformers trained on multiple Othello variants share a common board-state representation rather than isolating world models.

Reasoning

The paper introduces a controlled suite of Othello variants to study how transformers organize multiple world models, finding shared representations and transferable linear probes. Strengths include a clear experimental design and novel insights, but weaknesses are the limited scope (synthetic Othello games) and lack of real-world validation.

Read-first score

Read-first score 36.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 34.

Recency 6%

100

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

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=analysis,experiment

Topical relevance 29%

48.6

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

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 12%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 162.

Keyword Scores

world model

world dynamics prediction

generative world model

world simulator

interactive world model

model-based reinforcement learning world model

video world model

Deep Analysis

Innovations

Introduction of MetaOthello, a controlled suite of Othello variants with shared syntax but different rules or tokenizations
Discovery that transformers trained on mixed-variant data do not partition capacity into isolated sub-models but converge on a mostly shared board-state representation that transfers causally across variants
Demonstration that linear probes trained on one variant can intervene on another's internal state with effectiveness approaching matched probes
Finding that for isomorphic games with token remapping, representations are equivalent up to a single orthogonal rotation that generalizes across layers
Observation that when rules partially overlap, early layers maintain game-agnostic representations, a middle layer identifies game identity, and later layers specialize

Methodology

The authors introduce MetaOthello, a controlled suite of Othello variants with shared syntax but different rules or tokenizations. They train small GPTs on mixed-variant data and use linear probes and intervention analysis to study how multiple world models are organized in a shared representation space.

Key Results

Transformers trained on mixed-game data converge on a mostly shared board-state representation that transfers causally across variants; linear probes from one variant intervene on another with near-matched effectiveness. Isomorphic games yield representations equivalent via a single orthogonal rotation, and partially overlapping rules lead to early game-agnostic layers, a middle identity layer, and later specialized layers.

Resources

Paper