DEL-ToM: Inference-Time Scaling for Theory-of-Mind Reasoning via Dynamic Epistemic Logic

Theory-of-Mind (ToM) tasks pose a unique challenge for small language models (SLMs) with limited scale, which often lack the capacity to perform deep social reasoning. In this work, we propose DEL-ToM, a framework that improves ToM reasoning through inference-time scaling rather than architectural changes. Our approach decomposes ToM tasks into a sequence of belief updates grounded in Dynamic Epistemic Logic (DEL), enabling structured and transparent reasoning. We train a verifier, called the Process Belief Model (PBM), to score each belief update step using labels generated automatically via a DEL simulator. During inference, candidate belief traces generated by a language model are evaluated by the PBM, and the highest-scoring trace is selected. This allows SLMs to emulate more deliberate reasoning by allocating additional compute at test time. Experiments across multiple model scales and benchmarks show that DEL-ToM consistently improves performance, demonstrating that verifiable belief supervision can significantly enhance ToM abilities of SLMs without retraining.

SATBench: Benchmarking LLMs' Logical Reasoning via Automated Puzzle Generation from SAT Formulas

We introduce SATBench, a benchmark for evaluating the logical reasoning capabilities of large language models (LLMs) through logical puzzles derived from Boolean satisfiability (SAT) problems. Unlike prior work that focuses on inference rule-based reasoning, which often involves deducing conclusions from a set of premises, our approach leverages the search-based nature of SAT problems, where the objective is to find a solution that fulfills a specified set of logical constraints. Each instance in SATBench is generated from a SAT formula, then translated into a story context and conditions using LLMs. The generation process is fully automated and allows for adjustable difficulty by varying the number of clauses. All 2100 puzzles are validated through both LLM-assisted and solver-based consistency checks, with human validation on a subset. Experimental results show that even the strongest model, o4-mini, achieves only 65.0% accuracy on hard UNSAT problems, close to the random baseline of 50%. SATBench exposes fundamental limitations in the search-based logical reasoning abilities of current LLMs and provides a scalable testbed for future research in logical reasoning.

LangCoop: Collaborative Driving with Language

Multi-agent collaboration holds great promise for enhancing the safety, reliability, and mobility of autonomous driving systems by enabling information sharing among multiple connected agents. However, existing multi-agent communication approaches are hindered by limitations of existing communication media, including high bandwidth demands, agent heterogeneity, and information loss. To address these challenges, we introduce LangCoop, a new paradigm for collaborative autonomous driving that leverages natural language as a compact yet expressive medium for inter-agent communication. LangCoop features two key innovations: Mixture Model Modular Chain-of-thought (M$^3$CoT) for structured zero-shot vision-language reasoning and Natural Language Information Packaging (LangPack) for efficiently packaging information into concise, language-based messages. Through extensive experiments conducted in the CARLA simulations, we demonstrate that LangCoop achieves a remarkable 96\% reduction in communication bandwidth (< 2KB per message) compared to image-based communication, while maintaining competitive driving performance in the closed-loop evaluation. Our project page and code are at https://xiangbogaobarry.github.io/LangCoop/.

Sensitivity Meets Sparsity: The Impact of Extremely Sparse Parameter Patterns on Theory-of-Mind of Large Language Models

This paper investigates the emergence of Theory-of-Mind (ToM) capabilities in large language models (LLMs) from a mechanistic perspective, focusing on the role of extremely sparse parameter patterns. We introduce a novel method to identify ToM-sensitive parameters and reveal that perturbing as little as 0.001% of these parameters significantly degrades ToM performance while also impairing contextual localization and language understanding. To understand this effect, we analyze their interaction with core architectural components of LLMs. Our findings demonstrate that these sensitive parameters are closely linked to the positional encoding module, particularly in models using Rotary Position Embedding (RoPE), where perturbations disrupt dominant-frequency activations critical for contextual processing. Furthermore, we show that perturbing ToM-sensitive parameters affects LLM's attention mechanism by modulating the angle between queries and keys under positional encoding. These insights provide a deeper understanding of how LLMs acquire social reasoning abilities, bridging AI interpretability with cognitive science. Our results have implications for enhancing model alignment, mitigating biases, and improving AI systems designed for human interaction.

CodeARC: Benchmarking Reasoning Capabilities of LLM Agents for Inductive Program Synthesis

Inductive program synthesis, or programming by example, requires synthesizing functions from input-output examples that generalize to unseen inputs. While large language model agents have shown promise in programming tasks guided by natural language, their ability to perform inductive program synthesis is underexplored. Existing evaluation protocols rely on static sets of examples and held-out tests, offering no feedback when synthesized functions are incorrect and failing to reflect real-world scenarios such as reverse engineering. We propose CodeARC, the Code Abstraction and Reasoning Challenge, a new evaluation framework where agents interact with a hidden target function by querying it with new inputs, synthesizing candidate functions, and iteratively refining their solutions using a differential testing oracle. This interactive setting encourages agents to perform function calls and self-correction based on feedback. We construct the first large-scale benchmark for general-purpose inductive program synthesis, featuring 1114 functions. Among 18 models evaluated, o3-mini performs best with a success rate of 52.7%, highlighting the difficulty of this task. Fine-tuning LLaMA-3.1-8B-Instruct on curated synthesis traces yields up to a 31% relative performance gain. CodeARC provides a more realistic and challenging testbed for evaluating LLM-based program synthesis and inductive reasoning.

Competent but Rigid: Identifying the Gap in Empowering AI to Participate Equally in Group Decision-Making

Existing research on human-AI collaborative decision-making focuses mainly on the interaction between AI and individual decision-makers. There is a limited understanding of how AI may perform in group decision-making. This paper presents a wizard-of-oz study in which two participants and an AI form a committee to rank three English essays. One novelty of our study is that we adopt a speculative design by endowing AI equal power to humans in group decision-making.We enable the AI to discuss and vote equally with other human members. We find that although the voice of AI is considered valuable, AI still plays a secondary role in the group because it cannot fully follow the dynamics of the discussion and make progressive contributions. Moreover, the divergent opinions of our participants regarding an "equal AI" shed light on the possible future of human-AI relations.