We find that the ability of LLMs to maintain consistency varies across different moral scenarios.

Multimodal Large Language Models (MLLMs) demonstrate strong capabilities in handling image-text inputs. A common way to assess this ability is through multiple-choice Visual Question Answering (VQA). Earlier works have already revealed that these benchmarks are sensitive to answer choice order, a limitation that can be mitigated through careful design. Yet, we highlight additional, unexplored biases in prompt formatting that question the reliability of current MLLM evaluations. Specifically, we identify three key variation factors in prompt formatting and analyze their impact through a large-scale study involving $\mathbf{\text{seven}}$ MLLMs and $\mathbf{\text{five}}$ VQA datasets, spanning $\mathbf{48}$ distinct $\mathbf{\text{prompt format variations}}$. Our findings reveal that multiple-choice VQA is highly sensitive to minor prompt format changes, even when these changes are semantically neutral. We further demonstrate that these biases persist independently of known order biases or the MLLM's confidence in the correct answer. Finally, we demonstrate that existing bias mitigation strategies fail to address these newly identified biases.

Large language models (LLMs) are increasingly evaluated on single-answer multiple-choice tasks, yet many real-world problems require identifying all correct answers from a set of options. This capability remains underexplored. We introduce SATA-BENCH, the first dedicated benchmark for evaluating LLMs on Select All That Apply (SATA) questions across diverse domains, including reading comprehension, law, and biomedicine. Our evaluation of 27 open-source and proprietary models reveals a significant gap: even the strongest model achieves only 41.8% exact match, exposing LLMs' inability to reliably identify all correct answers. We find that this weakness stems from two core challenges: selection bias - models favor certain choices regardless of content, and count bias - models fail to predict the correct number of answers. To address these issues, we propose Choice Funnel, a decoding strategy that combines token debiasing with adaptive thresholding to guide models toward complete and accurate selections. Choice Funnel achieves up to 29% higher exact match than competitive baselines while reducing inference cost by over 64%. Our findings expose fundamental limitations in current LLMs and introduce a new framework for diagnosing and improving multi-answer reasoning. We release SATA-BENCH and Choice Funnel to promote LLM development for robust decision-making in realistic, multi-answer applications.

We find that the ability of LLMs to maintain consistency varies across different moral scenarios.

Multiple choice questions (MCQs) are a popular and important task for evaluating large language models (LLMs). Based on common strategies people use when answering MCQs, the process of elimination (PoE) has been proposed as an effective problem-solving method. Existing methods to the PoE generally fall into two categories: one involves having the LLM directly select the incorrect options, while the other involves scoring the options. However, both methods incur high computational costs and often perform worse than methods that directly answer the MCQs with the option IDs. To address this issue, this paper proposes a PoE based on option ID. Specifically, our method eliminates option by selecting the option ID with the lowest probability. We conduct experiments with 10 different LLMs in zero-shot settings on 7 publicly available datasets. The experimental results demonstrate that our method significantly improves the LLM's performance. Further analysis reveals that the sequential elimination strategy can effectively enhance the LLM's reasoning ability. Additionally, we find that sequential elimination is also applicable to few-shot settings and can be combined with debias methods to further improve LLM's performance.

In this paper, we introduce SailCompass, a reproducible and robust evaluation benchmark for assessing Large Language Models (LLMs) on Southeast Asian Languages (SEA). SailCompass encompasses three main SEA languages, eight primary tasks including 14 datasets covering three task types (generation, multiple-choice questions, and classification). To improve the robustness of the evaluation approach, we explore different prompt configurations for multiple-choice questions and leverage calibrations to improve the faithfulness of classification tasks. With SailCompass, we derive the following findings: (1) SEA-specialized LLMs still outperform general LLMs, although the gap has narrowed; (2) A balanced language distribution is important for developing better SEA-specialized LLMs; (3) Advanced prompting techniques (e.g., calibration, perplexity-based ranking) are necessary to better utilize LLMs. All datasets and evaluation scripts are public.

Multiple choice questions (MCQs) serve as a common yet important task format in the research of large language models (LLMs). This work shows that LLMs are vulnerable to option position changes in MCQs due to their inherent "selection bias", namely, they prefer to select specific option IDs as answers (like "Option A"). Through extensive empirical analyses with 20 LLMs on three benchmarks, we pinpoint that this behavioral bias primarily stems from LLMs' token bias, where the model a priori assigns more probabilistic mass to specific option ID tokens (e.g., A/B/C/D) when predicting answers from the option IDs. To mitigate selection bias, we propose a label-free, inference-time debiasing method, called PriDe, which separates the model's prior bias for option IDs from the overall prediction distribution. PriDe first estimates the prior by permutating option contents on a small number of test samples, which is then applied to debias the subsequent samples. We demonstrate that PriDe achieves superior debiasing effectiveness and computational efficiency to strong baselines. Furthermore, the prior estimated by PriDe is interpretable and can generalize well across different domains, highlighting its practical potential in broader scenarios. Current LLM-centric scenarios have widely utilized the task format of MCQ, for instance, Figure 1: A multiple choice question within benchmarks targeted at assessing LLMs (Hendrycks (MCQ) example from MMLU. In any scenario, we always expect LLMs to robustly select reliable answers in MCQs.