SpatiaLab: Can Vision-Language Models Perform Spatial Reasoning in the Wild?

Spatial reasoning is a fundamental aspect of human cognition, yet it remains a major challenge for contemporary vision-language models (VLMs). Prior work largely relied on synthetic or LLM-generated environments with limited task designs and puzzle-like setups, failing to capture the real-world complexity, visual noise, and diverse spatial relationships that VLMs encounter. To address this, we introduce SpatiaLab, a comprehensive benchmark for evaluating VLMs' spatial reasoning in realistic, unconstrained contexts. SpatiaLab comprises 1,400 visual question-answer pairs across six major categories: Relative Positioning, Depth & Occlusion, Orientation, Size & Scale, Spatial Navigation, and 3D Geometry, each with five subcategories, yielding 30 distinct task types. Each subcategory contains at least 25 questions, and each main category includes at least 200 questions, supporting both multiple-choice and open-ended evaluation. Experiments across diverse state-of-the-art VLMs, including open- and closed-source models, reasoning-focused, and specialized spatial reasoning models, reveal a substantial gap in spatial reasoning capabilities compared with humans. In the multiple-choice setup, InternVL3.5-72B achieves 54.93% accuracy versus 87.57% for humans. In the open-ended setting, all models show a performance drop of around 10-25%, with GPT-5-mini scoring highest at 40.93% versus 64.93% for humans. These results highlight key limitations in handling complex spatial relationships, depth perception, navigation, and 3D geometry. By providing a diverse, real-world evaluation framework, SpatiaLab exposes critical challenges and opportunities for advancing VLMs' spatial reasoning, offering a benchmark to guide future research toward robust, human-aligned spatial understanding. SpatiaLab is available at: https://spatialab-reasoning.github.io/.

Do I Really Know? Learning Factual Self-Verification for Hallucination Reduction

Factual hallucination remains a central challenge for large language models (LLMs). Existing mitigation approaches primarily rely on either external post-hoc verification or mapping uncertainty directly to abstention during fine-tuning, often resulting in overly conservative behavior. We propose VeriFY, a training-time framework that teaches LLMs to reason about factual uncertainty through consistency-based self-verification. VeriFY augments training with structured verification traces that guide the model to produce an initial answer, generate and answer a probing verification query, issue a consistency judgment, and then decide whether to answer or abstain. To address the risk of reinforcing hallucinated content when training on augmented traces, we introduce a stage-level loss masking approach that excludes hallucinated answer stages from the training objective while preserving supervision over verification behavior. Across multiple model families and scales, VeriFY reduces factual hallucination rates by 9.7 to 53.3 percent, with only modest reductions in recall (0.4 to 5.7 percent), and generalizes across datasets when trained on a single source. The source code, training data, and trained model checkpoints will be released upon acceptance.

There Is More to Refusal in Large Language Models than a Single Direction

Prior work argues that refusal in large language models is mediated by a single activation-space direction, enabling effective steering and ablation. We show that this account is incomplete. Across eleven categories of refusal and non-compliance, including safety, incomplete or unsupported requests, anthropomorphization, and over-refusal, we find that these refusal behaviors correspond to geometrically distinct directions in activation space. Yet despite this diversity, linear steering along any refusal-related direction produces nearly identical refusal to over-refusal trade-offs, acting as a shared one-dimensional control knob. The primary effect of different directions is not whether the model refuses, but how it refuses.

Exploring Alignment in Shared Cross-lingual Spaces

Despite their remarkable ability to capture linguistic nuances across diverse languages, questions persist regarding the degree of alignment between languages in multilingual embeddings. Drawing inspiration from research on high-dimensional representations in neural language models, we employ clustering to uncover latent concepts within multilingual models. Our analysis focuses on quantifying the \textit{alignment} and \textit{overlap} of these concepts across various languages within the latent space. To this end, we introduce two metrics \CA{} and \CO{} aimed at quantifying these aspects, enabling a deeper exploration of multilingual embeddings. Our study encompasses three multilingual models (\texttt{mT5}, \texttt{mBERT}, and \texttt{XLM-R}) and three downstream tasks (Machine Translation, Named Entity Recognition, and Sentiment Analysis). Key findings from our analysis include: i) deeper layers in the network demonstrate increased cross-lingual \textit{alignment} due to the presence of language-agnostic concepts, ii) fine-tuning of the models enhances \textit{alignment} within the latent space, and iii) such task-specific calibration helps in explaining the emergence of zero-shot capabilities in the models.\footnote{The code is available at \url{https://github.com/baselmousi/multilingual-latent-concepts}}

Improving Language Models Trained with Translated Data via Continual Pre-Training and Dictionary Learning Analysis

Training LLMs in low resources languages usually utilizes data augmentation with machine translation (MT) from English language. However, translation brings a number of challenges: there are large costs attached to translating and curating huge amounts of content with high-end machine translation solutions, the translated content carries over cultural biases, and if the translation is not faithful and accurate, the quality of the data degrades causing issues in the trained model. In this work we investigate the role of translation and synthetic data in training language models. We translate TinyStories, a dataset of 2.2M short stories for 3-4 year old children, from English to Arabic using the free NLLB-3B MT model. We train a number of story generation models of sizes 1M-33M parameters using this data. We identify a number of quality and task-specific issues in the resulting models. To rectify these issues, we further pre-train the models with a small dataset of synthesized high-quality stories, representing 1\% of the original training data, using a capable LLM in Arabic. We show using GPT-4 as a judge and dictionary learning analysis from mechanistic interpretability that the suggested approach is a practical means to resolve some of the translation pitfalls. We illustrate the improvement through case studies of linguistic issues and cultural bias.

Analyzing Multilingual Competency of LLMs in Multi-Turn Instruction Following: A Case Study of Arabic

While significant progress has been made in benchmarking Large Language Models (LLMs) across various tasks, there is a lack of comprehensive evaluation of their abilities in responding to multi-turn instructions in less-commonly tested languages like Arabic. Our paper offers a detailed examination of the proficiency of open LLMs in such scenarios in Arabic. Utilizing a customized Arabic translation of the MT-Bench benchmark suite, we employ GPT-4 as a uniform evaluator for both English and Arabic queries to assess and compare the performance of the LLMs on various open-ended tasks. Our findings reveal variations in model responses on different task categories, e.g., logic vs. literacy, when instructed in English or Arabic. We find that fine-tuned base models using multilingual and multi-turn datasets could be competitive to models trained from scratch on multilingual data. Finally, we hypothesize that an ensemble of small, open LLMs could perform competitively to proprietary LLMs on the benchmark.

Scaled-up Discovery of Latent Concepts in Deep NLP Models

Pre-trained language models (pLMs) learn intricate patterns and contextual dependencies via unsupervised learning on vast text data, driving breakthroughs across NLP tasks. Despite these achievements, these models remain black boxes, necessitating research into understanding their decision-making processes. Recent studies explore representation analysis by clustering latent spaces within pre-trained models. However, these approaches are limited in terms of scalability and the scope of interpretation because of high computation costs of clustering algorithms. This study focuses on comparing clustering algorithms for the purpose of scaling encoded concept discovery of representations from pLMs. Specifically, we compare three algorithms in their capacity to unveil the encoded concepts through their alignment to human-defined ontologies: Agglomerative Hierarchical Clustering, Leaders Algorithm, and K-Means Clustering. Our results show that K-Means has the potential to scale to very large datasets, allowing rich latent concept discovery, both on the word and phrase level.

LLMeBench: A Flexible Framework for Accelerating LLMs Benchmarking

The recent development and success of Large Language Models (LLMs) necessitate an evaluation of their performance across diverse NLP tasks in different languages. Although several frameworks have been developed and made publicly available, their customization capabilities for specific tasks and datasets are often complex for different users. In this study, we introduce the LLMeBench framework. Initially developed to evaluate Arabic NLP tasks using OpenAI's GPT and BLOOM models; it can be seamlessly customized for any NLP task and model, regardless of language. The framework also features zero- and few-shot learning settings. A new custom dataset can be added in less than 10 minutes, and users can use their own model API keys to evaluate the task at hand. The developed framework has been already tested on 31 unique NLP tasks using 53 publicly available datasets within 90 experimental setups, involving approximately 296K data points. We plan to open-source the framework for the community (https://github.com/qcri/LLMeBench/). A video demonstrating the framework is available online (https://youtu.be/FkQn4UjYA0s).

Benchmarking Arabic AI with Large Language Models

With large Foundation Models (FMs), language technologies (AI in general) are entering a new paradigm: eliminating the need for developing large-scale task-specific datasets and supporting a variety of tasks through set-ups ranging from zero-shot to few-shot learning. However, understanding FMs capabilities requires a systematic benchmarking effort by comparing FMs performance with the state-of-the-art (SOTA) task-specific models. With that goal, past work focused on the English language and included a few efforts with multiple languages. Our study contributes to ongoing research by evaluating FMs performance for standard Arabic NLP and Speech processing, including a range of tasks from sequence tagging to content classification across diverse domains. We start with zero-shot learning using GPT-3.5-turbo, Whisper, and USM, addressing 33 unique tasks using 59 publicly available datasets resulting in 96 test setups. For a few tasks, FMs performs on par or exceeds the performance of the SOTA models but for the majority it under-performs. Given the importance of prompt for the FMs performance, we discuss our prompt strategies in detail and elaborate on our findings. Our future work on Arabic AI will explore few-shot prompting, expand the range of tasks, and investigate additional open-source models.

DiPA: Diverse and Probabilistically Accurate Interactive Prediction

Accurate prediction is important for operating an autonomous vehicle in interactive scenarios. Previous interactive predictors have used closest-mode evaluations, which test if one of a set of predictions covers the ground-truth, but not if additional unlikely predictions are made. The presence of unlikely predictions can interfere with planning, by indicating conflict with the ego plan when it is not likely to occur. Closest-mode evaluations are not sufficient for showing a predictor is useful, an effective predictor also needs to accurately estimate mode probabilities, and to be evaluated using probabilistic measures. These two evaluation approaches, eg. predicted-mode RMS and minADE/FDE, are analogous to precision and recall in binary classification, and there is a challenging trade-off between prediction strategies for each. We present DiPA, a method for producing diverse predictions while also capturing accurate probabilistic estimates. DiPA uses a flexible representation that captures interactions in widely varying road topologies, and uses a novel training regime for a Gaussian Mixture Model that supports diversity of predicted modes, along with accurate spatial distribution and mode probability estimates. DiPA achieves state-of-the-art performance on INTERACTION and NGSIM, and improves over a baseline (MFP) when both closest-mode and probabilistic evaluations are used at the same time.