A common solution for mitigating outdated or incorrect information in Large Language Models (LLMs) is to provide updated facts in-context or through knowledge editing. However, these methods introduce knowledge conflicts when the knowledge update fails to overwrite the model’s parametric knowledge, which propagate to faulty reasoning. Current benchmarks for this problem, however, largely focus only on single knowledge updates and fact recall without evaluating how these updates affect downstream reasoning. In this work, we introduce Tʀᴀᴄᴋ (*Testing Reasoning Amid Conflicting Knowledge*), a new benchmark for studying how LLMs propagate new knowledge through multi-step reasoning when it conflicts with the model’s initial parametric knowledge. Spanning three reasoning-intensive scenarios (WIKI, CODE, and MATH), Tʀᴀᴄᴋ introduces multiple, realistic conflicts to mirror real-world complexity. Our results on Tʀᴀᴄᴋ reveal that providing updated facts to models for reasoning can worsen performance compared to providing no updated facts to a model, and that this performance degradation exacerbates as more updated facts are provided. We show this failure stems from both inability to faithfully integrate updated facts, but also flawed reasoning even when knowledge is integrated. Tʀᴀᴄᴋ provides a rigorous new benchmark to measure and guide future progress on propagating conflicting knowledge in multi-step reasoning.

Existing multi-agent learning approaches explicitly foster collaboration among Large Language Models (LLMs) to build stronger multi-agent systems (MAS), yet they still rely on re-executing the MAS during inference. This contrasts with human cognition, wherein individuals can internalize insights from interactions to improve later independent reasoning. To investigate whether multi-agent interaction can enhance LLMs’ independent problem-solving ability, we propose ILR (Interactive Learning for LLM Reasoning), a co-learning framework that integrates Dynamic Interaction and Perception Calibration. Dynamic Interaction adaptively selects cooperative or competitive strategies based on question difficulty and model capability, after which LLMs exchange information via Idea3 framework (Idea Sharing, Idea Analysis, and Idea Fusion), an interaction paradigm simulating human discussion, before producing final answers. Perception Calibration employs Group Relative Policy Optimization (GRPO) while integrating one LLM’s reward characteristics into another’s to strengthen interaction cohesion. We evaluate the effectiveness of ILR across three LLMs from two model families of varying scales on five mathematical and one coding benchmarks. We further investigate the advantage of Dynamic Interaction (i.e., boosting the robustness of stronger LLMs and surpassing pure strategy), and the scalability of ILR beyond two-model interactions.

Chain-of-Thought (CoT) has significantly enhanced LLM reasoning, yet often incurs substantial computational overhead due to “overthinking”: generating excessively long rationales without commensurate accuracy gains. Existing efficiency methods typically apply uniform compression, which overlooks a critical observation that reasoning complexity is heterogeneous at two distinct granularities: across different problems and within individual reasoning steps. This motivates our principle of Thinking Economically: intelligently allocating computational resources based on intrinsic task and step demands rather than pursuing uniform brevity. We propose Hierarchical Adaptive Budgeter (HAB), a training framework that operationalizes this principle through coarse-to-fine budgeting. At the inter-step level, HAB predicts the optimal reasoning depth for each problem. At the intra-step level, HAB learns step-specific token budgeting signals from PPL-derived step comparisons and an adaptive Pareto optimization objective that captures the local quality-efficiency trade-off, while a Fisher Information-based pruner further provides fine-grained training-time guidance, thereby encouraging the generator to internalize more economical reasoning patterns. Experiments on GSM8K and MATH500 show that HAB not only surpasses standard CoT in accuracy but also reduces token usage, achieving a stronger performance-efficiency trade-off than the compared baselines.

This paper presents our contribution to the Streamlining Discharge Documentation shared task organized as part of the ACL’24 workshop. We propose MEDISCHARGE (Meditron-7B Based Medical Summary Generation System for Discharge Me), an LLM-based system to generate Brief Hospital Course and Discharge Instruction summaries based on a patient’s Electronic Health Record. Our system is build on a Meditron-7B with context window extension, ensuring the system can handle cases of variable lengths with high quality. When the length of the input exceeds the system input limitation, we use a dynamic information selection framework to automatically extract important sections from the full discharge text. Then, extracted sections are removed in increasing order of importance until the input length requirement is met. We demonstrate our approach outperforms tripling the size of the context window of the model. Our system obtains a 0.289 overall score in the leaderboard, an improvement of 183% compared to the baseline, and a ROUGE-1 score of 0.444, achieving a second place performance in the shared task.