Large language models exhibit substantial performance variation across languages, even when solving semantically equivalent tasks. Existing analyses often treat this phenomenon as an observational disparity caused by differences in pretraining data, tokenization, or benchmark coverage. We study a complementary hypothesis: high-resource languages (HRLs) may more reliably elicit latent computations useful for task-specific (i.e. mathematical) reasoning, while lower-resource languages (LRLs) may under-activate those computations despite expressing the same task. To test this hypothesis, we introduce a mechanistic intervention framework for identifying and transferring task-relevant sparse latent features across languages. Using sparse autoencoders over residual-stream activations, we isolate features enriched in successful HRL task-specific reasoning while filtering out source-language and generic-generation features. We then construct steering directions from these features and inject them during LRL inference. The resulting interventions test whether the selected features are functionally involved in the observed reasoning gap: suppressing them should impair source-language reasoning, while activating them should partially recover target-language reasoning beyond random and non-task controls. Our framework reframes some cross-lingual reasoning gaps as failures of mechanism elicitation rather than capability absence, and offers a causally testable route to feature-mediated transfer without translation, fine-tuning, or changing the user-facing language.

Chain-of-thought prompting improves reasoning by generating intermediate steps as discrete text, but each step is forced to commit to a single token even when the model is uncertain. Continuous latent reasoning offers an alternative: a latent state can preserve multiple candidate interpretations without immediately selecting one token. Existing training-free approaches either place latent states only before decoding or switch between explicit and latent modes without optimizing the inserted state itself. We propose MidThink, a training-free framework that inserts and optimizes latent blocks both at the start of reasoning and adaptively during generation. When a confidence trigger detects sustained mid-stream uncertainty, MidThink inserts a new latent block and refines it with a rollout-based reward computed from short future continuations. We further show that initialization strongly affects whether a mid-stream block preserves a superposition over multiple tokens or collapses toward a single token. Across five mathematical reasoning benchmarks, MidThink improves over training-free latent-reasoning baselines, with the largest gains on the AIME benchmarks. Mechanistic analysis links these gains to three measurable properties of the latent block: distributional entropy across layers, attention centroid position, and stability under reinforcement learning updates.

Large language models (LLM) hold significant potential for applications in biomedicine, but they struggle with hallucinations and outdated knowledge.While retrieval-augmented generation (RAG) is generally employed to address these issues, it also has its own set of challenges: (1) LLMs are vulnerable to irrelevant or unhelpful context, (2) medical queries are often not well-targeted for helpful information, and (3) retrievers are prone to bias toward the specific source corpus they were trained on. In this study, we present RAG² (RAtionale-Guided RAG), a new framework for enhancing the reliability of RAG in biomedical contexts. RAG² incorporates three key innovations: a small filtering model trained on perplexity-based labels of rationales, which selectively augments informative snippets of documents while filtering out distractors; LLM-generated rationales as queries to improve the utility of retrieved snippets; a structure designed to retrieve snippets evenly from a comprehensive set of four biomedical corpora, effectively mitigating retriever bias. Our experiments demonstrate that RAG² improves the state-of-the-art LLMs of varying sizes, with improvements of up to 6.1%, and it outperforms the previous best medical RAG model by up to 5.6% across three medical question-answering benchmarks. Our code is available at https://github.com/dmis-lab/RAG2

Step-by-step reasoning has become a standard approach for large language models (LLMs) to tackle complex tasks. While this paradigm has proven effective, it raises a fundamental question: How can we verify that an LLM’s reasoning is accurately grounded in knowledge? To address this question, we introduce a novel evaluation suite that systematically assesses the knowledge grounding of intermediate reasoning. Our framework comprises three key components. (1) Principal Knowledge Collection, a large-scale repository of atomic knowledge essential for reasoning. Based on the collection, we propose (2) knowledge-grounded evaluation metrics designed to measure how well models recall and apply prerequisite knowledge in reasoning. These metrics are computed by our (3) evaluator LLM, a lightweight model optimized for cost-effective and reliable metric computation. Our evaluation suite demonstrates remarkable effectiveness in identifying missing or misapplied knowledge elements, providing crucial insights for uncovering fundamental reasoning deficiencies in LLMs. Beyond evaluation, we demonstrate how these metrics can be integrated into preference optimization, showcasing further applications of knowledge-grounded evaluation. Our evaluation suite is publicly available.

The increasing utilization of patient portals has amplified clinicians’ workloads, primarily due to the necessity of addressing detailed patient inquiries related to their health concerns. The ArchEHR-QA 2025 shared task aims to alleviate this burden by automatically generating accurate, evidence-grounded responses to patients’ questions based on their Electronic Health Records (EHRs). This paper presents a six-stage multi-agent framework specifically developed to identify essential clinical sentences for answering patient questions, leveraging large language models (LLMs). Our approach begins with OpenAI’s o3 model generating focused medical context to guide downstream reasoning. In the subsequent stages, GPT-4.1-based agents assess the relevance of individual sentences, recruit domain experts, and consolidate their judgments to identify essential information for constructing coherent, evidence-grounded responses. Our framework achieved an Overall Factuality score of 62.0 and an Overall Relevance Score of 52.9 on the development set, and corresponding scores of 58.6 and 48.8, respectively, on the test set.

Extreme activation outliers in Large Language Models (LLMs) critically degrade quantization performance, hindering efficient on-device deployment. While channel-wise operations and adaptive gradient scaling are recognized causes, practical mitigation remains challenging. We introduce Outlier-Safe Pre-Training (OSP), a practical guideline that proactively prevents outlier formation, rather than relying on post-hoc mitigation. OSP combines three key innovations: (1) the Muon optimizer, eliminating privileged bases while maintaining training efficiency, (2) Single-Scale RMSNorm, preventing channel-wise amplification, and (3) a learnable embedding projection, redistributing activation magnitudes. We validate OSP by training a 1.4B-parameter model on 1 trillion tokens, which is the first production-scale LLM trained without such outliers. Under aggressive 4-bit quantization, our OSP model achieves a 35.7 average score across 10 benchmarks (versus 26.5 for an Adam-trained model), with only a 2% training overhead. Remarkably, OSP models exhibit near-zero excess kurtosis (0.04) compared to extreme values (1818.56) in standard models, fundamentally altering LLM quantization behavior. Our work demonstrates that outliers are not inherent to LLMs but are consequences of training strategies, paving the way for more efficient LLM deployment.

Retrieval-augmented generation supports language models to strengthen their factual groundings by providing external contexts. However, language models often face challenges when given extensive information, diminishing their effectiveness in solving questions. Context compression tackles this issue by filtering out irrelevant information, but current methods still struggle in realistic scenarios where crucial information cannot be captured with a single-step approach. To overcome this limitation, we introduce CompAct, a novel framework that employs an active strategy to condense extensive documents without losing key information. Our experiments demonstrate that CompAct brings significant improvements in both performance and compression rate on multi-hop question-answering benchmarks. CompAct flexibly operates as a cost-efficient plug-in module with various off-the-shelf retrievers or readers, achieving exceptionally high compression rates (47x).

Recent advancements in large language models (LM) like OpenAI’s GPT-4 have shown promise in healthcare, particularly in medical question answering and clinical applications. However, their deployment raises privacy concerns and their size limits use in resource-constrained environments.Smaller open-source LMs have emerged as alternatives, but their reliability in medicine remains underexplored.This study evaluates small LMs in the medical field using the MEDIQA-CORR 2024 task, which assesses the ability of models to identify and correct errors in clinical notes. Initially, zero-shot inference and simple fine-tuning of small models resulted in poor performance. When fine-tuning with chain-of-thought (CoT) reasoning using synthetic data generated by GPT-4, their performance significantly improved. Meerkat-7B, a small LM trained with medical CoT reasoning, demonstrated notable performance gains. Our model outperforms other small non-commercial LMs and some larger models, achieving a 73.36 aggregate score on MEDIQA-CORR 2024.