FLEx: Language Modeling with Few-shot Language Explanations

Language models have become effective at a wide range of tasks, from math problem solving to open-domain question answering. However, they still make mistakes, and these mistakes are often repeated across related queries. Natural language explanations can help correct these errors, but collecting them at scale may be infeasible, particularly in domains where expert annotators are required. To address this issue, we introduce FLEx ($\textbf{F}$ew-shot $\textbf{L}$anguage $\textbf{Ex}$planations), a method for improving model behavior using a small number of explanatory examples. FLEx selects representative model errors using embedding-based clustering, verifies that the associated explanations correct those errors, and summarizes them into a prompt prefix that is prepended at inference-time. This summary guides the model to avoid similar errors on new inputs, without modifying model weights. We evaluate FLEx on CounterBench, GSM8K, and ReasonIF. We find that FLEx consistently outperforms chain-of-thought (CoT) prompting across all three datasets and reduces up to 83\% of CoT's remaining errors.

From Chains to Graphs: Self-Structured Reasoning for General-Domain LLMs

Large Language Models (LLMs) show strong reasoning ability in open-domain question answering, yet their reasoning processes are typically linear and often logically inconsistent. In contrast, real-world reasoning requires integrating multiple premises and solving subproblems in parallel. Existing methods, such as Chain-of-Thought (CoT), express reasoning in a linear textual form, which may appear coherent but frequently leads to inconsistent conclusions. Recent approaches rely on externally provided graphs and do not explore how LLMs can construct and use their own graph-structured reasoning, particularly in open-domain QA. To fill this gap, we novelly explore graph-structured reasoning of LLMs in general-domain question answering. We propose Self-Graph Reasoning (SGR), a framework that enables LLMs to explicitly represent their reasoning process as a structured graph before producing the final answer. We further construct a graph-structured reasoning dataset that merges multiple candidate reasoning graphs into refined graph structures for model training. Experiments on five QA benchmarks across both general and specialized domains show that SGR consistently improves reasoning consistency and yields a 17.74% gain over the base model. The LLaMA-3.3-70B model fine-tuned with SGR performs comparably to GPT-4o and surpasses Claude-3.5-Haiku, demonstrating the effectiveness of graph-structured reasoning.

AM$^3$Safety: Towards Data Efficient Alignment of Multi-modal Multi-turn Safety for MLLMs

Multi-modal Large Language Models (MLLMs) are increasingly deployed in interactive applications. However, their safety vulnerabilities become pronounced in multi-turn multi-modal scenarios, where harmful intent can be gradually reconstructed across turns, and security protocols fade into oblivion as the conversation progresses. Existing Reinforcement Learning from Human Feedback (RLHF) alignment methods are largely developed for single-turn visual question-answer (VQA) task and often require costly manual preference annotations, limiting their effectiveness and scalability in dialogues. To address this challenge, we present InterSafe-V, an open-source multi-modal dialogue dataset containing 11,270 dialogues and 500 specially designed refusal VQA samples. This dataset, constructed through interaction between several models, is designed to more accurately reflect real-world scenarios and includes specialized VQA pairs tailored for specific domains. Building on this dataset, we propose AM$^3$Safety, a framework that combines a cold-start refusal phase with Group Relative Policy Optimization (GRPO) fine-tuning using turn-aware dual-objective rewards across entire dialogues. Experiments on Qwen2.5-VL-7B-Instruct and LLaVA-NeXT-7B show more than 10\% decrease in Attack Success Rate (ASR) together with an increment of at least 8\% in harmless dimension and over 13\% in helpful dimension of MLLMs on multi-modal multi-turn safety benchmarks, while preserving their general abilities.

VietMed-MCQ: A Consistency-Filtered Data Synthesis Framework for Vietnamese Traditional Medicine Evaluation

Large Language Models (LLMs) have demonstrated remarkable proficiency in general medical domains. However, their performance significantly degrades in specialized, culturally specific domains such as Vietnamese Traditional Medicine (VTM), primarily due to the scarcity of high-quality, structured benchmarks. In this paper, we introduce VietMed-MCQ, a novel multiple-choice question dataset generated via a Retrieval-Augmented Generation (RAG) pipeline with an automated consistency check mechanism. Unlike previous synthetic datasets, our framework incorporates a dual-model validation approach to ensure reasoning consistency through independent answer verification, though the substring-based evidence checking has known limitations. The complete dataset of 3,190 questions spans three difficulty levels and underwent validation by one medical expert and four students, achieving 94.2 percent approval with substantial inter-rater agreement (Fleiss' kappa = 0.82). We benchmark seven open-source models on VietMed-MCQ. Results reveal that general-purpose models with strong Chinese priors outperform Vietnamese-centric models, highlighting cross-lingual conceptual transfer, while all models still struggle with complex diagnostic reasoning. Our code and dataset are publicly available to foster research in low-resource medical domains.

MARVEL: A Multi Agent-based Research Validator and Enabler using Large Language Models

We present MARVEL (https://ligogpt.mit.edu/marvel), a locally deployable, open-source framework for domain-aware question answering and assisted scientific research. It is designed to address the increasing demands of a digital assistant for scientific groups that can read highly technical data, cite precisely, and operate within authenticated networks. MARVEL combines a fast path for straightforward queries with a more deliberate DeepSearch mode that integrates retrieval-augmented generation and Monte Carlo Tree Search. It explores complementary subqueries, allocates more compute to promising branches, and maintains a global evidence ledger that preserves sources during drafting. We applied this framework in the context of gravitational-wave research related to the Laser Interferometer Gravitational-wave Observatory. Answers are grounded in a curated semantic index of research literature, doctoral theses, LIGO documents, and long-running detector electronic logbooks, with targeted web searches when appropriate. Because direct benchmarking against commercial LLMs cannot be performed on private data, we evaluated MARVEL on two publicly available surrogate datasets that capture comparable semantic and technical characteristics. On these benchmarks, MARVEL matches a GPT-4o mini baseline on literature-centric queries and substantially outperforms it on detector-operations content, where domain retrieval and guided reasoning are decisive. By making the complete framework and evaluation datasets openly available, we aim to provide a reproducible foundation for developing domain-specific scientific assistants.

pdfQA: Diverse, Challenging, and Realistic Question Answering over PDFs

PDFs are the second-most used document type on the internet (after HTML). Yet, existing QA datasets commonly start from text sources or only address specific domains. In this paper, we present pdfQA, a multi-domain 2K human-annotated (real-pdfQA) and 2K synthetic dataset (syn-pdfQA) differentiating QA pairs in ten complexity dimensions (e.g., file type, source modality, source position, answer type). We apply and evaluate quality and difficulty filters on both datasets, obtaining valid and challenging QA pairs. We answer the questions with open-source LLMs, revealing existing challenges that correlate with our complexity dimensions. pdfQA presents a basis for end-to-end QA pipeline evaluation, testing diverse skill sets and local optimizations (e.g., in information retrieval or parsing).

FormationEval, an open multiple-choice benchmark for petroleum geoscience

This paper presents FormationEval, an open multiple-choice question benchmark for evaluating language models on petroleum geoscience and subsurface disciplines. The dataset contains 505 questions across seven domains including petrophysics, petroleum geology and reservoir engineering, derived from three authoritative sources using a reasoning model with detailed instructions and a concept-based approach that avoids verbatim copying of copyrighted text. Each question includes source metadata to support traceability and audit. The evaluation covers 72 models from major providers including OpenAI, Anthropic, Google, Meta and open-weight alternatives. The top performers achieve over 97\% accuracy, with Gemini 3 Pro Preview reaching 99.8\%, while tier and domain gaps persist. Among open-weight models, GLM-4.7 leads at 98.6\%, with several DeepSeek, Llama, Qwen and Mistral models also exceeding 93\%. The performance gap between open-weight and closed models is narrower than expected, with several lower-cost open-weight models exceeding 90\% accuracy. Petrophysics emerges as the most challenging domain across all models, while smaller models show wider performance variance. Residual length bias in the dataset (correct answers tend to be longer) is documented along with bias mitigation strategies applied during construction. The benchmark, evaluation code and results are publicly available.

Reasoning Over Recall: Evaluating the Efficacy of Generalist Architectures vs. Specialized Fine-Tunes in RAG-Based Mental Health Dialogue Systems

The deployment of Large Language Models (LLMs) in mental health counseling faces the dual challenges of hallucinations and lack of empathy. While the former may be mitigated by RAG (retrieval-augmented generation) by anchoring answers in trusted clinical sources, there remains an open question as to whether the most effective model under this paradigm would be one that is fine-tuned on mental health data, or a more general and powerful model that succeeds purely on the basis of reasoning. In this paper, we perform a direct comparison by running four open-source models through the same RAG pipeline using ChromaDB: two generalist reasoners (Qwen2.5-3B and Phi-3-Mini) and two domain-specific fine-tunes (MentalHealthBot-7B and TherapyBot-7B). We use an LLM-as-a-Judge framework to automate evaluation over 50 turns. We find a clear trend: the generalist models outperform the domain-specific ones in empathy (3.72 vs. 3.26, $p < 0.001$) in spite of being much smaller (3B vs. 7B), and all models perform well in terms of safety, but the generalist models show better contextual understanding and are less prone to overfitting as we observe in the domain-specific models. Overall, our results indicate that for RAG-based therapy systems, strong reasoning is more important than training on mental health-specific vocabulary; i.e. a well-reasoned general model would provide more empathetic and balanced support than a larger narrowly fine-tuned model, so long as the answer is already grounded in clinical evidence.

AdaGReS:Adaptive Greedy Context Selection via Redundancy-Aware Scoring for Token-Budgeted RAG

Retrieval-augmented generation (RAG) is highly sensitive to the quality of selected context, yet standard top-k retrieval often returns redundant or near-duplicate chunks that waste token budget and degrade downstream generation. We present AdaGReS, a redundancy-aware context selection framework for token-budgeted RAG that optimizes a set-level objective combining query-chunk relevance and intra-set redundancy penalties. AdaGReS performs greedy selection under a token-budget constraint using marginal gains derived from the objective, and introduces a closed-form, instance-adaptive calibration of the relevance-redundancy trade-off parameter to eliminate manual tuning and adapt to candidate-pool statistics and budget limits. We further provide a theoretical analysis showing that the proposed objective exhibits epsilon-approximate submodularity under practical embedding similarity conditions, yielding near-optimality guarantees for greedy selection. Experiments on open-domain question answering (Natural Questions) and a high-redundancy biomedical (drug) corpus demonstrate consistent improvements in redundancy control and context quality, translating to better end-to-end answer quality and robustness across settings.

Retrieval Augmented Question Answering: When Should LLMs Admit Ignorance?

The success of expanded context windows in Large Language Models (LLMs) has driven increased use of broader context in retrieval-augmented generation. We investigate the use of LLMs for retrieval augmented question answering. While longer contexts make it easier to incorporate targeted knowledge, they introduce more irrelevant information that hinders the model's generation process and degrades its performance. To address the issue, we design an adaptive prompting strategy which involves splitting the retrieved information into smaller chunks and sequentially prompting a LLM to answer the question using each chunk. Adjusting the chunk size allows a trade-off between incorporating relevant information and reducing irrelevant information. Experimental results on three open-domain question answering datasets demonstrate that the adaptive strategy matches the performance of standard prompting while using fewer tokens. Our analysis reveals that when encountering insufficient information, the LLM often generates incorrect answers instead of declining to respond, which constitutes a major source of error. This finding highlights the need for further research into enhancing LLMs' ability to effectively decline requests when faced with inadequate information.