The paper investigates how children build mental models under uncertainty using a Box Task that involves interacting with a stochastic, partially observed environment. It introduces a Bayesian particle-based inference framework that maintains a small set of working hypotheses and continually generates new ones, offering two implementations: constraint-based and programmatic. The study demonstrates that accounting for subjective reliability of evidence and open-ended hypothesis spaces is essential to explain children's behavior, and shows that programmatic hypotheses can be generated via LLM-based program synthesis, robust across different prompts and LLM backends.
arXiv:2605. 24528v2 Announce Type: replace Abstract: Real world decision-making requires constructing mental models under uncertainty over evidence, over the underlying causal rules, and over the state of the world itself.
The paper introduces a computational model that encodes symbolic knowledge as mental programs combining natural language and source code, and uses LLM-guided Bayesian learning to sequentially infer these programs. It demonstrates that this approach satisfies data‑efficiency, uncertainty handling, and flexibility, reproducing human inductive learning and active inquiry behaviors such as anchoring and garden‑pathing. In contrast, pure LLMs and classic Bayesian models either fail the task, do not match human behavior, or require prohibitive computational resources.
The paper introduces PIP-ToM, a framework that addresses the evidence-admissibility problem in multimodal Theory of Mind (ToM) by ensuring that only evidence accessible to an agent and diagnostic of its mental state is used in hypothesis evaluation. PIP-ToM builds agent‑specific evidence stores, aligns them with structured mental‑state claims, and calibrates Bayesian inference scores while preserving the underlying inference process. Using GPT‑4o, the method outperforms existing ToM approaches on MuMA‑ToM, MMToM‑QA, and Hi‑ToM benchmarks, achieving accuracies of 92.56%, 95.00%, and 95.00% respectively.
arXiv:2606. 19353v1 Announce Type: cross Abstract: In-Context Learning (ICL) allows LLMs to adapt to new tasks from a few demonstrations, but its reliability remains a concern: predictions are highly sensitive to both prompt design and the model's ability to understand the context, obscuring whether failures arise from data properties or model limitations.
The paper investigates whether large language models (LLMs) can perform principled Bayesian inference when updating beliefs in context. By creating an evaluation framework that includes qualitative epistemological tests, quantitative benchmarks, and analyses of downstream impacts, the authors find that LLMs typically violate Bayes' rule but that their performance improves dramatically when equipped with advanced reasoning and code‑execution tools, allowing them to approximate analytically intractable posteriors. The study suggests that appropriately tool‑augmented LLMs could already be leveraged to accelerate scientific discovery through complex belief updates.
arXiv:2606. 30850v1 Announce Type: new Abstract: Large language models (LLMs) are typically deployed in multi-turn conversations, where each turn provides new evidence that should reduce epistemic uncertainty about their environment.
The paper proposes a distributional theory explaining how large language models (LLMs) incorporate external evidence into their decision-making process. It identifies three key predictions: (1) evidence is more persuasive when it aligns with the model’s prior beliefs, (2) models more readily accept errors from their own internal processes than from external sources, and (3) the same evidence can improve weaker models while harming stronger ones. Extensive experiments across ten million trials, twelve LLMs from four families, and eight domains—including quantum mechanics, physics, genetics, and molecular biology—confirm these predictions and reveal that evidence integration occurs late in the network as a structured sequence of steps rather than through a simple trust metric.
arXiv:2610.35609v1 Announce Type: cross Abstract: Modeling the latent, multi-step reasoning processes of learners is a core challenge for adaptive tutoring systems, which often rely on simple right/w...
arXiv:2607. 03847v1 Announce Type: new Abstract: Understanding why discovered scenarios become critical in scenario-based testing is essential for effectively leveraging them in decision-making systems.
arXiv:2606. 03969v1 Announce Type: cross Abstract: Reliable uncertainty communication is critical to the trustworthiness of LLMs, yet faithful calibration (FC)--the alignment between models' intrinsic and (linguistically) expressed confidence--is a persistent failure mode.
The paper introduces LLM-as-an-Investigator, an evidence-first approach for interactive problem diagnosis that mitigates premature commitment to incorrect hypotheses. It implements this through the Solution Investigator Agent (SIA), which coordinates a Diagnostic Reasoner, a Hypothesis Investigator, and a Counterfactual Investigator to gather discriminative evidence and refine competing explanations. The method is evaluated on a structured corpus of 303 technical forum threads, outperforming several baseline strategies and demonstrating robustness to misleading user hypotheses.
The paper investigates whether the text of chain‑of‑thought reasoning steps actually reflects their true importance for a model’s final answer. By defining step importance as the advantage in expected reward when a step is included, the authors use Monte Carlo rollouts to estimate ground truth and then test whether large language model judges can identify high‑advantage steps. They find that capable LLMs can beat a prevalence baseline but still fall far short of a noise ceiling, and that fine‑tuning a step‑level critic improves detection for incorrect responses but remains distant from the ceiling for correct ones, indicating that step importance is only partially recoverable from the reasoning trace text.