# Simulated Annealing Enhances Theory-of-Mind Reasoning in Autoregressive Language Models

arXiv: https://arxiv.org/abs/2601.12269 · HTML: https://mlsanity.com/paper/2601.12269/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 18 of its 35 references that are in the map.
- First posted on arXiv 18 Jan 2026.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

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

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Tree of Thoughts: Deliberate Problem Solving with Large Language Models](https://mlsanity.com/paper/2305.10601/) (2023)
- [World Models](https://mlsanity.com/paper/1803.10122/) (2018)
- [The Reversal Curse: LLMs trained on "A is B" fail to learn "B is A"](https://mlsanity.com/paper/2309.12288/) (2023)
- [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)
- [Recursive Introspection: Teaching Language Model Agents How to Self-Improve](https://mlsanity.com/paper/2407.18219/) (2024)
- [From Word Models to World Models: Translating from Natural Language to the Probabilistic Language of Thought](https://mlsanity.com/paper/2306.12672/) (2023)
- [Reasoning with Sampling: Your Base Model is Smarter Than You Think](https://mlsanity.com/paper/2510.14901/) (2025)
- [How FaR Are Large Language Models From Agents with Theory-of-Mind?](https://mlsanity.com/paper/2310.03051/) (2023)
- [Incoherent Probability Judgments in Large Language Models](https://mlsanity.com/paper/2401.16646/) (2024)
- [Re-evaluating Theory of Mind evaluation in large language models](https://mlsanity.com/paper/2502.21098/) (2025)
- [Do LLMs Exhibit Human-Like Reasoning? Evaluating Theory of Mind in LLMs for Open-Ended Responses](https://mlsanity.com/paper/2406.05659/) (2024)
- [Formalizing Neurath's Ship: Approximate Algorithms for Online Causal Learning](https://mlsanity.com/paper/1609.04212/) (2016)
- [What Should Embeddings Embed? Autoregressive Models Represent Latent Generating Distributions](https://mlsanity.com/paper/2406.03707/) (2024)

## Built on by (most cited first, 2 of 2)

- [Large Language Models Can Take False First Steps at Inference-time Planning](https://mlsanity.com/paper/2602.02991/) (2026)
- [Understanding Human-like Solutions in Combinatorial Optimization via Learning and Search](https://mlsanity.com/paper/2607.23854/) (2026)

## Abstract (arXiv)

Autoregressive language models are next-token predictors and have been criticized for only optimizing surface plausibility (i.e., local coherence) rather than maintaining correct latent-state representations (i.e., global coherence). Because Theory of Mind (ToM) tasks crucially depend on reasoning about latent mental states of oneself and others, such models are therefore often thought to fail at ToM. While post-training methods can improve ToM performance, we show that strong ToM capability can be recovered directly from the base model without any additional weight updates or verifications. Our approach builds on recent power-sampling methods (Karan & Du, 2025) that use Markov chain Monte Carlo (MCMC) to sample from sharpened sequence-level (rather than token-level) probability distributions of autoregressive language models. We further find that incorporating annealing, where the tempered distribution is gradually shifted from high to low temperature, substantially improves ToM performance over fixed-temperature power sampling. Together, these results suggest that sampling-based optimization provides a powerful way to extract latent capabilities from language models without retraining.