Understanding the interplay between intra-modality dependencies (the contribution of an individual modality to a target task) and inter-modality dependencies (the relationships between modalities and the target task) is fundamental to advancing multi-modal learning. However, the nature of and interaction between these dependencies within current benchmark evaluations remains poorly characterized. In this work, we present a large-scale empirical study to quantify these dependencies across 23 visual question-answering benchmarks using multi-modal large language models (MLLMs) covering domains such as general and expert knowledge reasoning, optical character recognition, and document understanding. Our findings show that the reliance on vision, question (text), and their interaction varies significantly, both across and within benchmarks. We discover that numerous benchmarks intended to mitigate text-only biases have inadvertently amplified image-only dependencies. This characterization persists across model sizes, as larger models often use these intra-modality dependencies to achieve high performance that mask an underlying lack of multi-modal reasoning. We provide a quantitative characterization of multi-modal datasets, enabling a principled approach to multi-modal benchmark design and evaluation.

Item difficulty plays a crucial role in test performance, interpretability of scores, and equity for all test-takers, especially in large-scale assessments. Traditional approaches to item difficulty modeling rely on field testing and classical test theory (CTT)-based item analysis or item response theory (IRT) calibration, which can be time-consuming and costly. To overcome these challenges, text-based approaches leveraging machine learning and language models, have emerged as promising alternatives. This paper reviews and synthesizes 37 articles on automated item difficulty prediction in large-scale assessment settings published through May 2025. For each study, we delineate the dataset, difficulty parameter, subject domain, item type, number of items, training and test data split, input, features, model, evaluation criteria, and model performance outcomes. Results showed that although classic machine learning models remain relevant due to their interpretability, state-of-the-art language models, using both small and large transformer-based architectures, can capture syntactic and semantic patterns without the need for manual feature engineering. Uniquely, model performance outcomes were summarized to serve as a benchmark for future research and overall, text-based methods have the potential to predict item difficulty with root mean square error (RMSE) as low as 0.165, Pearson correlation as high as 0.87, and accuracy as high as 0.806. The review concludes by discussing implications for practice and outlining future research directions for automated item difficulty modeling.

The widespread adoption of AI models, especially foundation models (FMs), has made a profound impact on numerous domains. However, it also raises significant ethical concerns, including bias issues. Although numerous efforts have been made to quantify and mitigate social bias in AI models, geographic bias (in short, geo-bias) receives much less attention, which presents unique challenges. While previous work has explored ways to quantify geo-bias, these measures are model-specific (e.g., mean absolute deviation of LLM ratings) or spatially implicit (e.g., average fairness scores of all spatial partitions). We lack a model-agnostic, universally applicable, and spatially explicit geo-bias evaluation framework that allows researchers to fairly compare the geo-bias of different AI models and to understand what spatial factors contribute to the geo-bias. In this paper, we establish an information-theoretic framework for geo-bias evaluation, called GeoBS (Geo-Bias Scores). We demonstrate the generalizability of the proposed framework by showing how to interpret and analyze existing geo-bias measures under this framework. Then, we propose three novel geo-bias scores that explicitly take intricate spatial factors (multi-scalability, distance decay, and anisotropy) into consideration. Finally, we conduct extensive experiments on 3 tasks, 8 datasets, and 8 models to demonstrate that both task-specific GeoAI models and general-purpose foundation models may suffer from various types of geo-bias. This framework will not only advance the technical understanding of geographic bias but will also establish a foundation for integrating spatial fairness into the design, deployment, and evaluation of AI systems.

The parameter-efficient fine-tuning paradigm has garnered significant attention with the advancement of foundation models. Although numerous methods have been proposed to reduce the number of trainable parameters, their substantial memory overhead remains a critical bottleneck that hinders practical deployment. In this paper, we observe that model activations constitute a major source of memory consumption, especially under large batch sizes and long context lengths; however, the rank of the activations remains consistently low. Motivated by this insight, we propose a memory-efficient fine-tuning approach Low-Rank Activation Compression (LoRAct). Unlike prior work, LoRAct provides a more flexible and versatile compressing strategy that can be applied online during the forward pass without the need for any calibration data. Moreover, LoRAct incorporates a novel sampling-based orthogonal decomposition algorithm specifically designed for low-rank matrices, offering improved computational efficiency and a tighter error bound compared to the widely used RSVD. Experiments on both vision and language tasks demonstrate the effectiveness of LoRAct. Notably, LoRAct further reduces activation memory by approximately 80% in comparison with the widely adopted LoRA method, while maintaining competitive performance. The source code is available at https://github.com/shijxcs/meft.

Frame of Reference (FoR) is a fundamental concept in spatial reasoning that humans utilize to comprehend and describe space. With the rapid progress in Multimodal Language models, the moment has come to integrate this long-overlooked dimension into these models. In particular, in text-to-image (T2I) generation, even state-of-the-art models exhibit a significant performance gap when spatial descriptions are provided from perspectives other than the camera. To address this limitation, we propose Frame of Reference-guided Spatial Adjustment in LLM-based Diffusion Editing (FoR-SALE), an extension of the Self-correcting LLM-controlled Diffusion (SLD) framework for T2I. For-Sale evaluates the alignment between a given text and an initially generated image, and refines the image based on the Frame of Reference specified in the spatial expressions. It employs vision modules to extract the spatial configuration of the image, while simultaneously mapping the spatial expression to a corresponding camera perspective. This unified perspective enables direct evaluation of alignment between language and vision. When misalignment is detected, the required editing operations are generated and applied. FoR-SALE applies novel latent-space operations to adjust the facing direction and depth of the generated images. We evaluate FoR-SALE on two benchmarks specifically designed to assess spatial understanding with FoR. Our framework improves the performance of state-of-the-art T2I models by up to 5.3% using only a single round of correction.

Binary code similarity detection is a core task in reverse engineering. It supports malware analysis and vulnerability discovery by identifying semantically similar code in different contexts. Modern methods have progressed from manually engineered features to vector representations. Hand-crafted statistics (e.g., operation ratios) are interpretable, but shallow and fail to generalize. Embedding-based methods overcome this by learning robust cross-setting representations, but these representations are opaque vectors that prevent rapid verification. They also face a scalability-accuracy trade-off, since high-dimensional nearest-neighbor search requires approximations that reduce precision. Current approaches thus force a compromise between interpretability, generalizability, and scalability. We bridge these gaps using a language model-based agent to conduct structured reasoning analysis of assembly code and generate features such as input/output types, side effects, notable constants, and algorithmic intent. Unlike hand-crafted features, they are richer and adaptive. Unlike embeddings, they are human-readable, maintainable, and directly searchable with inverted or relational indexes. Without any matching training, our method respectively achieves 42% and 62% for recall@1 in cross-architecture and cross-optimization tasks, comparable to embedding methods with training (39% and 34%). Combined with embeddings, it significantly outperforms the state-of-the-art, demonstrating that accuracy, scalability, and interpretability can coexist.

The creation of high-quality multimodal datasets remains fundamental for advancing role-playing capabilities in large language models (LLMs). While existing works predominantly focus on text-based persona simulation, Audio Role-Playing (ARP) presents unique challenges due to the need for synchronized alignment of semantic content and vocal characteristics. To address this gap, we propose AudioRole, a meticulously curated dataset from 13 TV series spanning 1K+ hours with 1M+ character-grounded dialogues, providing synchronized audio-text pairs annotated with speaker identities and contextual metadata. In addition, to demonstrate the effectiveness of the dataset, we introduced ARP-Eval, a dual-aspect evaluation framework that assesses both response quality and role fidelity. Empirical validation showing GLM-4-Voice trained on AudioRole (which we called ARP-Model) achieve an average Acoustic Personalization score of 0.31, significantly outperforming the original GLM-4-voice and the more powerful model MiniCPM-O-2.6, which specifically supports role-playing in one-shot scenarios. The ARP-Model also achieves a Content Personalization score of 0.36, surpassing the untrained original model by about 38% and maintaining the same level as MiniCPM-O-2.6. AudioRole features dialogues from over 115 main characters, 6 trained ARP-Models that role-play different characters, and evaluation protocols. Together, they provide an essential resource for advancing audio-grounded role-playing research.

Communication challenges between autistic and neurotypical individuals stem from a mutual lack of understanding of each other's distinct, and often contrasting, communication styles. Yet, autistic individuals are expected to adapt to neurotypical norms, making interactions inauthentic and mentally exhausting for them. To help redress this imbalance, we build NeuroBridge, an online platform that utilizes large language models (LLMs) to simulate: (a) an AI character that is direct and literal, a style common among many autistic individuals, and (b) four cross-neurotype communication scenarios in a feedback-driven conversation between this character and a neurotypical user. Through NeuroBridge, neurotypical individuals gain a firsthand look at autistic communication, and reflect on their role in shaping cross-neurotype interactions. In a user study with 12 neurotypical participants, we find that NeuroBridge improved their understanding of how autistic people may interpret language differently, with all describing autism as a social difference that "needs understanding by others" after completing the simulation. Participants valued its personalized, interactive format and described AI-generated feedback as "constructive", "logical" and "non-judgmental". Most perceived the portrayal of autism in the simulation as accurate, suggesting that users may readily accept AI-generated (mis)representations of disabilities. To conclude, we discuss design implications for disability representation in AI, the need for making NeuroBridge more personalized, and LLMs' limitations in modeling complex social scenarios.

Real-world videos often show routine activities punctuated by memorable, surprising events. However, most Video-LLMs process videos by sampling frames uniformly, likely missing critical moments that define a video's narrative. We introduce SPIKE, an inference-time framework that quantifies Bayesian Surprise as the belief update triggered by new visual evidence in the video stream, identifying moments where new visual evidence conflicts with prior beliefs. SPIKE effectively localizes surprise in videos, strongly correlated with humans on positive (FunQA) and negative (Oops!) surprise benchmarks. Since the beliefs of zero-shot Video-LLMs are often suboptimal, we develop SPIKE-RL, which leverages GRPO to optimize belief hypotheses based on a reward signal from the video caption. SPIKE and SPIKE-RL guide query-agnostic surprise-weighted frame sampling, which allocates more frames to interesting moments in the video. With this strategy, we achieve consistent performance gains on five downstream benchmarks over uniform sampling. By enabling Video-LLMs to track beliefs and register surprise, our work paves the way for more robust models that can revise their understanding in response to new information.

Language models (LMs) encode substantial factual knowledge, but often produce answers judged as incorrect. We hypothesize that many of these answers are actually correct, but are expressed in alternative surface forms that are dismissed due to an overly strict evaluation, leading to an underestimation of models' parametric knowledge. We propose Retrieval-Constrained Decoding (RCD), a decoding strategy that restricts model outputs to unique surface forms. We introduce YAGO-QA, a dataset of 19,137 general knowledge questions. Evaluating open-source LMs from 135M to 70B parameters, we show that standard decoding undervalues their knowledge. For instance, Llama-3.1-70B scores only 32.3% F1 with vanilla decoding but 46.0% with RCD. Similarly, Llama-3.1-8B reaches 33.0% with RCD, outperforming the larger model under vanilla decoding. We publicly share the code and dataset at https://github.com/Rajjaa/disambiguated-LLM.