AI Paper Atlas

Research map / Large language models / Agents, code and math

SimpleToM: Exposing the Gap between Explicit ToM Inference and Implicit ToM Application in LLMs

Yuling Gu, Oyvind Tafjord, Hyunwoo Kim, Jared Moore, Ronan Le Bras, Peter Clark, Yejin Choi

ICLR 2026 · Large language models › Agents, code and math · approach: Mathematical and structural reasoning benchmarks · 53 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

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
ask behavior prediction instead of mental state question (applied vs explicit ToM)GPT-4o · SimpleToMaccuracy (%): 95.6 -> 49.5Hurt
ask judgment question instead of mental state questionLlama-3.1-405B · SimpleToMaccuracy (%): 97.8 -> 10.0Hurt
add generic chain-of-thought prompt (ablation)GPT-4o · SimpleToM behavior predictionaccuracy (%): 49.5 -> 62.8Improved
add generic chain-of-thought prompt (ablation)GPT-4o · SimpleToM judgmentaccuracy (%): 15.3 -> 39.2Improved
add mental state reminder to prompt (ablation)Llama-3.1-405B · SimpleToM behavior predictionaccuracy (%): 58.2 -> 89.5Improved
add mental state reminder to prompt (ablation)GPT-4o · SimpleToM judgmentaccuracy (%): 15.3 -> 42.2Improved
add mental state reminder to prompt (ablation)Claude-3.5-Sonnet · SimpleToM judgmentaccuracy (%): 24.9 -> 84.1Improved
add generic system prompt (SysP) about considering mental states (ablation)GPT-4o, Llama-3.1-405B, Claude-3.5-Sonnet · SimpleToM behavior predictionaccuracy change: -2.2% to +6.3%No clear effect
add explicit awareness hint to system prompt (SysP*) (ablation)GPT-4o, Llama-3.1-405B, Claude-3.5-Sonnet · SimpleToM behavior predictionaccuracy change: +19.1% to +25.1%Improved
use awareness-specific CoT* prompt instead of generic CoT (ablation)GPT-4o · SimpleToM judgmentaccuracy (%): 39.2 -> 86.7Improved
combine mental state reminder with CoT* (ablation)Claude-3.5-Sonnet · SimpleToMaverage accuracy (%): 63.3 -> 97.1Improved
use reasoning model (o1-preview) instead of non-reasoning modelso1-preview · SimpleToM behavior predictionaccuracy (%): 84.1Improved

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