# Developmental Trajectories of Situation Modeling and Mentalizing in Transformer Language Models

arXiv: https://arxiv.org/abs/2606.28524 · HTML: https://mlsanity.com/paper/2606.28524/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 18 of its 54 references that are in the map.
- First posted on arXiv 26 Jun 2026.

## Builds on (most cited first, 18 of 18)

- [Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling](https://mlsanity.com/paper/2304.01373/) (2023)
- [In-context Learning and Induction Heads](https://mlsanity.com/paper/2209.11895/) (2022)
- [AI and the Everything in the Whole Wide World Benchmark](https://mlsanity.com/paper/2111.15366/) (2021)
- [Large Language Models Fail on Trivial Alterations to Theory-of-Mind Tasks](https://mlsanity.com/paper/2302.08399/) (2023)
- [Evaluating Large Language Models in Theory of Mind Tasks](https://mlsanity.com/paper/2302.02083/) (2023)
- [Understanding Social Reasoning in Language Models with Language Models](https://mlsanity.com/paper/2306.15448/) (2023)
- [Clever Hans or Neural Theory of Mind? Stress Testing Social Reasoning in Large Language Models](https://mlsanity.com/paper/2305.14763/) (2023)
- [Sudden Drops in the Loss: Syntax Acquisition, Phase Transitions, and Simplicity Bias in MLMs](https://mlsanity.com/paper/2309.07311/) (2023)
- [Findings of the BabyLM Challenge: Sample-Efficient Pretraining on Developmentally Plausible Corpora](https://mlsanity.com/paper/2504.08165/) (2025)
- [LLM Circuit Analyses Are Consistent Across Training and Scale](https://mlsanity.com/paper/2407.10827/) (2024)
- [LLMs achieve adult human performance on higher-order theory of mind tasks](https://mlsanity.com/paper/2405.18870/) (2024)
- [Do Large Language Models know what humans know?](https://mlsanity.com/paper/2209.01515/) (2022)
- [Benchmarks as Microscopes: A Call for Model Metrology](https://mlsanity.com/paper/2407.16711/) (2024)
- [Language Model Behavioral Phases are Consistent Across Architecture, Training Data, and Scale](https://mlsanity.com/paper/2510.24963/) (2025)
- [Predicting the Emergence of Induction Heads in Language Model Pretraining](https://mlsanity.com/paper/2511.16893/) (2025)
- [Toward a Theory of Generalizability in LLM Mechanistic Interpretability Research](https://mlsanity.com/paper/2509.22831/) (2025)
- [Start Making Sense(s): A Developmental Probe of Attention Specialization Using Lexical Ambiguity](https://mlsanity.com/paper/2511.21974/) (2025)
- [Language Statistics and False Belief Reasoning: Evidence from 41 Open-Weight LMs](https://mlsanity.com/paper/2602.16085/) (2026)

## Abstract (arXiv)

Recent work suggests that Large Language Models (LLMs) are sensitive to the belief states of agents described by text, as measured by the false belief task (FBT), yet persistent concerns of construct validity remain. We adopt a **developmental perspective**, tracing the pattern of mental state reasoning behavior -- and likely **preconditions** for this behavior -- across multiple training stages in the Olmo2 and Pythia language model suites. We find that above-chance FBT performance depends both on model size and sufficient training volume, emerges relatively late in pretraining, and is most improved by post-training interventions (SFT, DPO) in the condition most diagnostic of mentalizing (False Belief, Implicit). However, FBT performance is fragile: consistent with past work, the use of non-factive verbs (e.g., thinks) increases false belief attributions even in the True Belief condition. To contextualize these findings, we track the emergence of **situation modeling**: the ability to report on basic factual properties of a described scene. Situation modeling accuracy generally precedes and exceeds FBT accuracy, yet situational representations also prove surprisingly incoherent in certain respects: when asked about the knowledge states of the Antagonist agent -- who always knows the item's true location -- Olmo2 13b is consistently influenced both by the Target agent's knowledge state and the presence of non-factive verbs. Together, these results suggest that larger, sufficiently trained models build partially coherent situation models in a developmentally appropriate sequence, yet display surprising fragility -- highlighting the value of developmental and stress-testing approaches for evaluating LLM capabilities.