Measuring Epistemic Humility in Multimodal Large Language Models

Hallucinations in multimodal large language models (MLLMs) -- where the model generates content inconsistent with the input image -- pose significant risks in real-world applications, from misinformation in visual question answering to unsafe errors in decision-making. Existing benchmarks primarily test recognition accuracy, i.e., evaluating whether models can select the correct answer among distractors. This overlooks an equally critical capability for trustworthy AI: recognizing when none of the provided options are correct, a behavior reflecting epistemic humility. We present HumbleBench, a new hallucination benchmark designed to evaluate MLLMs' ability to reject plausible but incorrect answers across three hallucination types: object, relation, and attribute. Built from a panoptic scene graph dataset, we leverage fine-grained scene graph annotations to extract ground-truth entities and relations, and prompt GPT-4-Turbo to generate multiple-choice questions, followed by a rigorous manual filtering process. Each question includes a "None of the above" option, requiring models not only to recognize correct visual information but also to identify when no provided answer is valid. We evaluate a variety of state-of-the-art MLLMs -- including both general-purpose and specialized reasoning models -- on HumbleBench and share valuable findings and insights with the community. By incorporating explicit false-option rejection, HumbleBench fills a key gap in current evaluation suites, providing a more realistic measure of MLLM reliability in safety-critical settings. Our code and dataset are released publicly and can be accessed at https://github.com/maifoundations/HumbleBench.

Automated Classification of Tutors' Dialogue Acts Using Generative AI: A Case Study Using the CIMA Corpus

This study explores the use of generative AI for automating the classification of tutors' Dialogue Acts (DAs), aiming to reduce the time and effort required by traditional manual coding. This case study uses the open-source CIMA corpus, in which tutors' responses are pre-annotated into four DA categories. Both GPT-3.5-turbo and GPT-4 models were tested using tailored prompts. Results show that GPT-4 achieved 80% accuracy, a weighted F1-score of 0.81, and a Cohen's Kappa of 0.74, surpassing baseline performance and indicating substantial agreement with human annotations. These findings suggest that generative AI has strong potential to provide an efficient and accessible approach to DA classification, with meaningful implications for educational dialogue analysis. The study also highlights the importance of task-specific label definitions and contextual information in enhancing the quality of automated annotation. Finally, it underscores the ethical considerations associated with the use of generative AI and the need for responsible and transparent research practices. The script of this research is publicly available at https://github.com/liqunhe27/Generative-AI-for-educational-dialogue-act-tagging.

Large Language Model-Based Automatic Formulation for Stochastic Optimization Models

This paper presents the first integrated systematic study on the performance of large language models (LLMs), specifically ChatGPT, to automatically formulate and solve stochastic optimiza- tion problems from natural language descriptions. Focusing on three key categories, joint chance- constrained models, individual chance-constrained models, and two-stage stochastic linear programs (SLP-2), we design several prompts that guide ChatGPT through structured tasks using chain-of- thought and modular reasoning. We introduce a novel soft scoring metric that evaluates the struc- tural quality and partial correctness of generated models, addressing the limitations of canonical and execution-based accuracy. Across a diverse set of stochastic problems, GPT-4-Turbo outperforms other models in partial score, variable matching, and objective accuracy, with cot_s_instructions and agentic emerging as the most effective prompting strategies. Our findings reveal that with well-engineered prompts and multi-agent collaboration, LLMs can facilitate specially stochastic formulations, paving the way for intelligent, language-driven modeling pipelines in stochastic opti- mization.

Posterior-GRPO: Rewarding Reasoning Processes in Code Generation

Reinforcement learning (RL) has significantly advanced code generation for large language models (LLMs). However, current paradigms rely on outcome-based rewards from test cases, neglecting the quality of the intermediate reasoning process. While supervising the reasoning process directly is a promising direction, it is highly susceptible to reward hacking, where the policy model learns to exploit the reasoning reward signal without improving final outcomes. To address this, we introduce a unified framework that can effectively incorporate the quality of the reasoning process during RL. First, to enable reasoning evaluation, we develop LCB-RB, a benchmark comprising preference pairs of superior and inferior reasoning processes. Second, to accurately score reasoning quality, we introduce an Optimized-Degraded based (OD-based) method for reward model training. This method generates high-quality preference pairs by systematically optimizing and degrading initial reasoning paths along curated dimensions of reasoning quality, such as factual accuracy, logical rigor, and coherence. A 7B parameter reward model with this method achieves state-of-the-art (SOTA) performance on LCB-RB and generalizes well to other benchmarks. Finally, we introduce Posterior-GRPO (P-GRPO), a novel RL method that conditions process-based rewards on task success. By selectively applying rewards to the reasoning processes of only successful outcomes, P-GRPO effectively mitigates reward hacking and aligns the model's internal reasoning with final code correctness. A 7B parameter model with P-GRPO achieves superior performance across diverse code generation tasks, outperforming outcome-only baselines by 4.5%, achieving comparable performance to GPT-4-Turbo. We further demonstrate the generalizability of our approach by extending it to mathematical tasks. Our models, dataset, and code are publicly available.

P-Aligner: Enabling Pre-Alignment of Language Models via Principled Instruction Synthesis

Large Language Models (LLMs) are expected to produce safe, helpful, and honest content during interaction with human users, but they frequently fail to align with such values when given flawed instructions, e.g., missing context, ambiguous directives, or inappropriate tone, leaving substantial room for improvement along multiple dimensions. A cost-effective yet high-impact way is to pre-align instructions before the model begins decoding. Existing approaches either rely on prohibitive test-time search costs or end-to-end model rewrite, which is powered by a customized training corpus with unclear objectives. In this work, we demonstrate that the goal of efficient and effective preference alignment can be achieved by P-Aligner, a lightweight module generating instructions that preserve the original intents while being expressed in a more human-preferred form. P-Aligner is trained on UltraPrompt, a new dataset synthesized via a proposed principle-guided pipeline using Monte-Carlo Tree Search, which systematically explores the space of candidate instructions that are closely tied to human preference. Experiments across different methods show that P-Aligner generally outperforms strong baselines across various models and benchmarks, including average win-rate gains of 28.35% and 8.69% on GPT-4-turbo and Gemma-2-SimPO, respectively. Further analyses validate its effectiveness and efficiency through multiple perspectives, including data quality, search strategies, iterative deployment, and time overhead.

Large Language Model-Based Framework for Explainable Cyberattack Detection in Automatic Generation Control Systems

The increasing digitization of smart grids has improved operational efficiency but also introduced new cybersecurity vulnerabilities, such as False Data Injection Attacks (FDIAs) targeting Automatic Generation Control (AGC) systems. While machine learning (ML) and deep learning (DL) models have shown promise in detecting such attacks, their opaque decision-making limits operator trust and real-world applicability. This paper proposes a hybrid framework that integrates lightweight ML-based attack detection with natural language explanations generated by Large Language Models (LLMs). Classifiers such as LightGBM achieve up to 95.13% attack detection accuracy with only 0.004 s inference latency. Upon detecting a cyberattack, the system invokes LLMs, including GPT-3.5 Turbo, GPT-4 Turbo, and GPT-4o mini, to generate human-readable explanation of the event. Evaluated on 100 test samples, GPT-4o mini with 20-shot prompting achieved 93% accuracy in identifying the attack target, a mean absolute error of 0.075 pu in estimating attack magnitude, and 2.19 seconds mean absolute error (MAE) in estimating attack onset. These results demonstrate that the proposed framework effectively balances real-time detection with interpretable, high-fidelity explanations, addressing a critical need for actionable AI in smart grid cybersecurity.

KnowMap: Efficient Knowledge-Driven Task Adaptation for LLMs

While Large Language Models (LLMs) possess significant capabilities in open-world agent tasks, they also face challenges in rapidly adapting to new, specialized tasks due to their reliance on static pre-trained knowledge. Traditional methods such as fine-tuning are often costly, data-intensive, and may lead to "catastrophic forgetting." Therefore, we present KnowMap, a novel approach that dynamically constructs a knowledge base from environmental and experiential data. KnowMap fine-tunes a small knowledge-embedding model to equip a larger LLM with valuable task-specific knowledge. Our experiments on the ScienceWorld benchmark demonstrate 17.71% improvement for the performance of gpt-4-turbo model. KnowMap not only provides an efficient and effective means for LLM task-adapting, but also highlights how integrating environmental and experiential knowledge can enhance LLMs' reasoning capabilities.

The impact of fine tuning in LLaMA on hallucinations for named entity extraction in legal documentation

The extraction of information about traffic accidents from legal documents is crucial for quantifying insurance company costs. Extracting entities such as percentages of physical and/or psychological disability and the involved compensation amounts is a challenging process, even for experts, due to the subtle arguments and reasoning in the court decision. A two-step procedure is proposed: first, segmenting the document identifying the most relevant segments, and then extracting the entities. For text segmentation, two methodologies are compared: a classic method based on regular expressions and a second approach that divides the document into blocks of n-tokens, which are then vectorized using multilingual models for semantic searches (text-embedding-ada-002/MiniLM-L12-v2 ). Subsequently, large language models (LLaMA-2 7b, 70b, LLaMA-3 8b, and GPT-4 Turbo) are applied with prompting to the selected segments for entity extraction. For the LLaMA models, fine-tuning is performed using LoRA. LLaMA-2 7b, even with zero temperature, shows a significant number of hallucinations in extractions which are an important contention point for named entity extraction. This work shows that these hallucinations are substantially reduced after finetuning the model. The performance of the methodology based on segment vectorization and subsequent use of LLMs significantly surpasses the classic method which achieves an accuracy of 39.5%. Among open-source models, LLaMA-2 70B with finetuning achieves the highest accuracy 79.4%, surpassing its base version 61.7%. Notably, the base LLaMA-3 8B model already performs comparably to the finetuned LLaMA-2 70B model, achieving 76.6%, highlighting the rapid progress in model development. Meanwhile, GPT-4 Turbo achieves the highest accuracy at 86.1%.

Probing Association Biases in LLM Moderation Over-Sensitivity

Large Language Models are widely used for content moderation but often misclassify benign comments as toxic, leading to over-sensitivity. While previous research attributes this issue primarily to the presence of offensive terms, we reveal a potential cause beyond token level: LLMs exhibit systematic topic biases in their implicit associations. Inspired by cognitive psychology's implicit association tests, we introduce Topic Association Analysis, a semantic-level approach to quantify how LLMs associate certain topics with toxicity. By prompting LLMs to generate free-form scenario imagination for misclassified benign comments and analyzing their topic amplification levels, we find that more advanced models (e.g., GPT-4 Turbo) demonstrate stronger topic stereotype despite lower overall false positive rates. These biases suggest that LLMs do not merely react to explicit, offensive language but rely on learned topic associations, shaping their moderation decisions. Our findings highlight the need for refinement beyond keyword-based filtering, providing insights into the underlying mechanisms driving LLM over-sensitivity.

Hydra: Structured Cross-Source Enhanced Large Language Model Reasoning

Retrieval-augmented generation (RAG) enhances large language models (LLMs) by incorporating external knowledge. Current hybrid RAG system retrieves evidence from both knowledge graphs (KGs) and text documents to support LLM reasoning. However, it faces challenges like handling multi-hop reasoning, multi-entity questions, multi-source verification, and effective graph utilization. To address these limitations, we present Hydra, a training-free framework that unifies graph topology, document semantics, and source reliability to support deep, faithful reasoning in LLMs. Hydra handles multi-hop and multi-entity problems through agent-driven exploration that combines structured and unstructured retrieval, increasing both diversity and precision of evidence. To tackle multi-source verification, Hydra uses a tri-factor cross-source verification (source trustworthiness assessment, cross-source corroboration, and entity-path alignment), to balance topic relevance with cross-modal agreement. By leveraging graph structure, Hydra fuses heterogeneous sources, guides efficient exploration, and prunes noise early. Comprehensive experiments on seven benchmark datasets show that Hydra achieves overall state-of-the-art results on all benchmarks with GPT-3.5, outperforming the strong hybrid baseline ToG-2 by an average of 20.3% and up to 30.1%. Furthermore, Hydra enables smaller models (e.g., Llama-3.1-8B) to achieve reasoning performance comparable to that of GPT-4-Turbo.