ABCD: All Biases Come Disguised

TL;DR

This work tackles biases in MCQ-based evaluation of large language models by introducing NonsenseQA as a diagnostic tool and a bias-reduced evaluation protocol, Matched-and-Dashed (M&D), which uses uniform dash labels and full-text answer generation with semantic matching. Across 13 open-source LLMs and five benchmarks (CSQA, ARC, GPQA, INCLUDE, MMLU-Pro), M&D substantially lowers accuracy variance across answer permutations while causing only a modest drop in mean performance, exposing genuine reasoning and multilingual capabilities. The study demonstrates the robustness and generalizability of the approach, including ablations showing insensitivity to sentence-similarity choices and resilience to label-design choices, with a modest computational overhead (~3%). These results imply more reliable cross-benchmark comparisons and a clearer separation between superficial shortcut use and true reasoning, informing safer and more credible AI evaluation practices.

Abstract

Multiple-choice question (MCQ) benchmarks have been a standard evaluation practice for measuring LLMs' ability to reason and answer knowledge-based questions. Through a synthetic NonsenseQA benchmark, we observe that different LLMs exhibit varying degrees of label-position-few-shot-prompt bias, where the model either uses the answer position, the label in front of the answer, the distributions of correct answers present in the few-shot prompt, or a combination of all to answer each MCQ question. We propose a simple bias-reduced evaluation protocol that replaces the labels of each question with uniform, unordered labels and prompts the LLM to use the whole answer presented. With a simple sentence similarity model, we demonstrate improved robustness and lower standard deviation between different permutations of answers with a minimal drop in LLM's performance, exposing the LLM's capabilities under reduced evaluation artifacts, without any help from the prompt examples or the option labels. Across multiple benchmarks and models, this protocol substantially improves the robustness to answer permutations, reducing mean accuracy variance $3\times$ with only a minimal decrease in the mean model's performance. Through ablation studies on various embedding models and similarity functions, we show that the method is more robust than the standard ones.

Figures (20)

• Figure 1: An example of the wrong conclusion reached by the Gemma-3-12 B-it model on one of the questions from the ARC clark2018arc dataset with our proposed method. By changing the presentations from standard evaluation (with different labels and predicting only the answer label) to matched (with uniform labels and predicting a whole answer, shown on the left), we eliminate the label bias present, and the model arrives at a correct conclusion that aligns with its reasoning.
• Figure 2: An example question from NonsenseQA with random answers and a golden answer chosen at random as "D. Arms".
• Figure 3: Comparison of the matched prediction with dashes as labels (M&D; our method) with standard letter prediction with letters (A/B/C/D) as labels (S&L) on NonsenseQA with a 5-shot prompt. The boxes illustrate the model performance under all possibilities of "answer-moving attacks", where the whiskers indicate the minimum and maximum accuracy for each model. Each dot represents the performance of the original permutations. Additionally, each star symbolizes a SCORE nalbandyan-etal-2025-score robustness metric.
• Figure 4: Comparison of the matched prediction with dashes as labels (M&D; our method) with standard letter prediction with letters as labels (S&L) on CSQA talmor-etal-2019-commonsenseqa with a 5-shot prompt. The boxes illustrate the model performance under all possibilities of ”answer-moving attacks”, where the whiskers indicate the minimum and maximum accuracy for each model. Each dot represents the performance of the original permutations. Additionally, each star symbolizes a SCORE nalbandyan-etal-2025-score robustness metric.
• Figure 5: Cross-benchmark Spearman and Kendall Tau rank correlation agreement