AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

MetaOthello: A Controlled Study of Multiple World Models in Transformers

Aviral Chawla, Galen Hall, Juniper L Lovato

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Transformer expressivity and learnability · 0 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Train on mixed-variant data instead of single game8-layer GPT · Othello variantsalpha score: about 0.5% dipNo clear effect
Use probe from one game to intervene on another (cross-probe) (ablation)Classic-NoMidFlip and Classic-DelFlank models · Board-state intervention errorprediction error: correct 0.2 vs cross 0.2 (Classic, Classic-NoMidFlip)No clear effect
Apply Procrustes rotation to Classic vs Iago probe weights (ablation)Classic-Iago mixed model · Probe weight cosine similaritycosine similarity: 0.03 -> 0.98Improved
Apply a single global orthogonal rotation to residual stream before predicting Iago moves (ablation)Classic-Iago mixed model · Classic-to-Iago transferalpha score: -2.9 -> near-perfectImproved
Add Procrustes alignment to Classic vs DelFlank probes (ablation)Classic-DelFlank mixed model · Probe weight cosine similaritycosine similarity: 0.67 -> >0.90Improved
Add game-ID steering vector at layer 5 (ablation)Classic-NoMidFlip model · Ambiguous prefixesNoMidFlip-consistent move shareImproved
Steer full disambiguation circuit (L4 MLP, L5 head 5, L5 MLP) (ablation)Classic-NoMidFlip model · Held-out ambiguous prefixesClassic share of unique-move mass: 0.48 -> 0.68 (Classic), 0.01 (NoMidFlip)Improved
Steer runner-up layer-5 head as control (ablation)Classic-NoMidFlip model · Held-out ambiguous prefixesClassic share of unique-move mass: stays at baselineNo clear effect
Ablate disambiguation circuit (ablation)Classic-NoMidFlip model · Held-out ambiguous prefixesvalid-move mass: 0.999 -> 0.964Hurt
Ablate disambiguation circuit (ablation)Classic-NoMidFlip model · Held-out ambiguous prefixesmass on wrong-game-only moves: 0.0001 -> 0.0115Hurt
Steer game identity at early layers 2-3 instead of layer 5 (ablation)Classic-DelFlank model · Ambiguous sequencesnormalized alpha toward DelFlank: delta alpha about 0.75 at low move numbers; layer 5 about 0Improved
Layer-2 steering toward DelFlank (ablation)Classic-DelFlank model · Layer-5 DelFlank probe accuracyprobe accuracy (%): 67.1 -> 77.4Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy