Spatial expressions in situated communication can be ambiguous, as their meanings vary depending on the frames of reference (FoR) adopted by speakers and listeners. While spatial language understanding and reasoning by vision-language models (VLMs) have gained increasing attention, potential ambiguities in these models are still under-explored. To address this issue, we present the COnsistent Multilingual Frame Of Reference Test (COMFORT), an evaluation protocol to systematically assess the spatial reasoning capabilities of VLMs. We evaluate nine state-of-the-art VLMs using COMFORT. Despite showing some alignment with English conventions in resolving ambiguities, our experiments reveal significant shortcomings of VLMs: notably, the models (1) exhibit poor robustness and consistency, (2) lack the flexibility to accommodate multiple FoRs, and (3) fail to adhere to languagespecific or culture-specific conventions in cross-lingual tests, as English tends to dominate other languages. With a growing effort to align vision-language models with human cognitive intuitions, we call for more attention to the ambiguous nature and cross-cultural diversity of spatial reasoning. The recent success of large language models has sparked breakthroughs in multi-modalities, leading to the development of many vision-language models (VLMs; Chen et al., 2023b; OpenAI, 2024; Reid et al., 2024, inter alia). With some benchmarks developed to evaluate the downstream performance of these models (Liu et al., 2023c; Yue et al., 2024), there has been growing excitement around evaluations and analyses inspired by human cognitive capabilities such as referential grounding (Ma et al., 2023a), compositional reasoning (Ma et al., 2023c), visual illusions (Zhang et al., 2023; Guan et al., 2024), and theory of mind (Jin et al., 2024). One direction among them that captures significant attention is spatial language understanding and reasoning, leading to several benchmarks (Kamath et al., 2023; Liu et al., 2023a) and enhanced models (Chen et al., 2024a; Cheng et al., 2024). Indeed, spatial cognition is a crucial part of human cognitive capability, developed since infancy and continuing through the elementary school years (Tommasi & Laeng, 2012; Vasilyeva & Lourenco, 2012). Language is closely intertwined with spatial cognition, with each contributing to the acquisition of the other (Hayward & Tarr, 1995; Regier & Carlson, 2001; Pyers et al., 2010; Pruden et al., 2011; Gentner et al., 2013). While spatial language and non-linguistic spatial representations in memory are closely correlated and share foundational properties, they are, to some extent, divergent-- spatial conventions are not consistently preserved across different languages or tasks, and humans demonstrate flexibility in using multiple coordinate systems for both non-linguistic reasoning and linguistic expressions (Munnich et al., 2001; Shusterman & Li, 2016).

The growing use of large language models (LLMs) has raised concerns regarding their safety. While many studies have focused on English, the safety of LLMs in Arabic, with its linguistic and cultural complexities, remains under-explored. Here, we aim to bridge this gap. In particular, we present an Arab-region-specific safety evaluation dataset consisting of 5,799 questions, including direct attacks, indirect attacks, and harmless requests with sensitive words, adapted to reflect the socio-cultural context of the Arab world. To uncover the impact of different stances in handling sensitive and controversial topics, we propose a dual-perspective evaluation framework. It assesses the LLM responses from both governmental and opposition viewpoints. Experiments over five leading Arabic-centric and multilingual LLMs reveal substantial disparities in their safety performance. This reinforces the need for culturally specific datasets to ensure the responsible deployment of LLMs.

Multi-step reasoning ability of large language models is crucial in tasks such as math and tool utilization. Current researches predominantly focus on enhancing model performance in these multi-step reasoning tasks through fine-tuning with Chain-of-Thought (CoT) steps, yet these methods tend to be heuristic, without exploring nor resolving the bottleneck. In this study, we subdivide CoT reasoning into two parts: arranging and executing, and identify that the bottleneck of models mainly lies in arranging rather than executing. Based on this finding, we propose a plan-based training and reasoning method that guides models to generate arranging steps through abstract plans. We experiment on both math (GSM8k) and tool utilization (ToolBench) benchmarks. Results show that compared to fine-tuning directly with CoT data, our approach achieves a better performance on alleviating arranging bottleneck, particularly excelling in long-distance reasoning generalization.

Following the advancement of large language models (LLMs), the development of LLM-based autonomous agents has become increasingly prevalent. As a result, the need to understand the security vulnerabilities of these agents has become a critical task. We examine how ReAct agents can be exploited using a straightforward yet effective method we refer to as the foot-in-the-door attack. Our experiments show that indirect prompt injection attacks, prompted by harmless and unrelated requests (such as basic calculations) can significantly increase the likelihood of the agent performing subsequent malicious actions. Our results show that once a ReAct agents thought includes a specific tool or action, the likelihood of executing this tool in the subsequent steps increases significantly, as the agent seldom re-evaluates its actions. Consequently, even random, harmless requests can establish a foot-in-the-door, allowing an attacker to embed malicious instructions into the agents thought process, making it more susceptible to harmful directives. To mitigate this vulnerability, we propose implementing a simple reflection mechanism that prompts the agent to reassess the safety of its actions during execution, which can help reduce the success of such attacks.

Despite the success of Large Language Models (LLMs) across various fields, their potential to generate untruthful, biased and harmful responses poses significant risks, particularly in critical applications. This highlights the urgent need for systematic methods to detect and prevent such misbehavior. Extensive experiments across various tasks and models reveal clear distinctions in the causal distributions between normal behavior and misbehavior, enabling the development of accurate, lightweight detectors for a variety of misbehavior detection tasks. Large language models (LLMs) demonstrate advanced capabilities in mimicking human language and styles for diverse applications (OpenAI, 2023), from literary creation (Yuan et al., 2022) to code generation (Li et al., 2023; Wang et al., 2023b). At the same time, they have shown the potential to misbehave in various ways, raising serious concerns about their use in critical real-world applications. First, LLMs can inadvertently produce untruthful responses, fabricating information that may be plausible but entirely fictitious, thus misleading users or misrepresenting facts (Rawte et al., 2023). Second, LLMs can be exploited for malicious purposes, such as through jailbreak attacks (Liu et al., 2024; Zou et al., 2023b; Zeng et al., 2024), where the model's safety mechanisms are bypassed to produce harmful outputs. Third, the generation of toxic responses such as insulting or offensive content remains a significant concern (Wang & Chang, 2022). Lastly, biased responses, which can appear as discriminatory or prejudiced remarks, are especially troubling as they have the potential to reinforce stereotypes and undermine societal efforts toward equality and inclusivity (Stanovsky et al., 2019; Zhao et al., 2018). Numerous attempts have been made to detect LLM misbehavior (Pacchiardi et al., 2023; Robey et al., 2024; Sap et al., 2020; Caselli et al., 2021). However, existing approaches often face two significant limitations. First, they tend to focus on a single type of misbehavior, which reduces the overall effectiveness of each method and requires the integration of multiple systems to comprehensively address the diverse forms of misbehavior. Second, many methods rely on analyzing the model's responses, which can be inefficient or even ineffective, particularly for longer outputs. Additionally, they are often vulnerable to adaptive adversarial attacks (Sato et al., 2018; Hartvigsen et al., 2022). As a result, there is an urgent need for more general and robust misbehavior detection methods capable of identifying (and mitigating) the full range of LLM misbehavior. In this work, we introduce LLMScan, an approach designed to address this critical need.

Accelerating research on Large Multimodal Models (LMMs) in non-English languages is crucial for enhancing user experiences across broader populations. In this paper, we introduce JMMMU (Japanese MMMU), the first large-scale Japanese benchmark designed to evaluate LMMs on expert-level tasks based on the Japanese cultural context. To facilitate comprehensive culture-aware evaluation, JMMMU features two complementary subsets: (i) culture-agnostic (CA) subset, where the culture-independent subjects (e.g., Math) are selected and translated into Japanese, enabling one-to-one comparison with its English counterpart MMMU; and (ii) culture-specific (CS) subset, comprising newly crafted subjects that reflect Japanese cultural context. Using the CA subset, we observe performance drop in many LMMs when evaluated in Japanese, which is purely attributable to language variation. Using the CS subset, we reveal their inadequate Japanese cultural understanding. Further, by combining both subsets, we identify that some LMMs perform well on the CA subset but not on the CS subset, exposing a shallow understanding of the Japanese language that lacks depth in cultural understanding. We hope this work will not only help advance LMM performance in Japanese but also serve as a guideline to create high-standard, culturally diverse benchmarks for multilingual LMM development. The project page is https://mmmu-japanese-benchmark.github.io/JMMMU/.

We propose a general pipeline to automate the extraction of labels from radiology reports using large language models, which we validate on spinal MRI reports. The efficacy of our labelling method is measured on five distinct conditions: spinal cancer, stenosis, spondylolisthesis, cauda equina compression and herniation. Using open-source models, our method equals or surpasses GPT-4 on a held-out set of reports. Furthermore, we show that the extracted labels can be used to train imaging models to classify the identified conditions in the accompanying MR scans. All classifiers trained using automated labels achieve comparable performance to models trained using scans manually annotated by clinicians. Code can be found at https://github.com/robinyjpark/AutoLabelClassifier.

Trustworthiness is an essential prerequisite for the real-world application of large language models. In this paper, we focus on the trustworthiness of language models with respect to retrieval augmentation. Despite being supported with external evidence, retrieval-augmented generation still suffers from hallucinations, one primary cause of which is the conflict between contextual and parametric knowledge. We deem that retrieval-augmented language models have the inherent capabilities of supplying response according to both contextual and parametric knowledge. Inspired by aligning language models with human preference, we take the first step towards aligning retrieval-augmented language models to a status where it responds relying merely on the external evidence and disregards the interference of parametric knowledge. Specifically, we propose a reinforcement learning based algorithm Trustworthy-Alignment, theoretically and experimentally demonstrating large language models' capability of reaching a trustworthy status without explicit supervision on how to respond. Our work highlights the potential of large language models on exploring its intrinsic abilities by its own and expands the application scenarios of alignment from fulfilling human preference to creating trustworthy agents.

We present a novel soft prompt based framework, SoftSRV, that leverages a frozen pre-trained large language model (LLM) to generate targeted synthetic text sequences. Given a sample from the target distribution, our proposed framework uses data-driven loss minimization to train a parameterized "contextual" soft prompt. This soft prompt is then used to steer the frozen LLM to generate synthetic sequences that are similar to the target distribution. We argue that SoftSRV provides a practical improvement over common hard-prompting approaches that rely on human-curated prompt-templates, which can be idiosyncratic, labor-intensive to craft, and may need to be specialized per domain. We empirically evaluate SoftSRV and hard-prompting baselines by generating synthetic data to fine-tune a small Gemma model on three different domains (coding, math, reasoning). To stress the generality of SoftSRV, we perform these evaluations without any particular specialization of the framework to each domain. We find that SoftSRV significantly improves upon hard-prompting baselines, generating data with superior fine-tuning performance and that better matches the target distribution according to the MAUVE similarity metric.

The quality is a crucial issue for crowd annotations. Answer aggregation is an important type of solution. The aggregated answers estimated from multiple crowd answers to the same instance are the eventually collected annotations, rather than the individual crowd answers themselves. Recently, the capability of Large Language Models (LLMs) on data annotation tasks has attracted interest from researchers. Most of the existing studies mainly focus on the average performance of individual crowd workers; several recent works studied the scenarios of aggregation on categorical labels and LLMs used as label creators. However, the scenario of aggregation on text answers and the role of LLMs as aggregators are not yet well-studied. In this paper, we investigate the capability of LLMs as aggregators in the scenario of close-ended crowd text answer aggregation. We propose a human-LLM hybrid text answer aggregation method with a Creator-Aggregator Multi-Stage (CAMS) crowdsourcing framework. We make the experiments based on public crowdsourcing datasets. The results show the effectiveness of our approach based on the collaboration of crowd workers and LLMs.