← Back to all news
arXiv Machine Learning By Fabio Rosenthal, Sebastian Schmidt, Thorsten Graf, Thorsten Bagdonat, Stephan G\"unnemann, Leo Schwinn

Unexplored flaws in multiple-choice VQA make benchmarking unreliable

Read the original on arXiv Machine Learning →

The paper demonstrates that multiple‑choice visual question answering (MC‑VQA) benchmarks are unreliable because model performance is highly sensitive to semantically neutral prompt formatting choices—such as option ID sets, delimiters, and separators—despite protocols that mitigate option‑order effects. Across seven multimodal large language models and five datasets, the authors observed frequent rank reversals when systematically varying 48 equivalent prompt formats, attributing the instability to tokenizer‑induced token fusion or removal and to how option ID sets influence attention patterns. Consequently, MC‑VQA rankings correlate weakly with open‑ended evaluation, revealing that MC‑VQA reflects option‑selection dynamics as well as multimodal reasoning.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.

Related stories

arXiv Machine Learning
Jun 25

Same Evidence, Different Answer: Auditing Order Sensitivity in Multimodal Large Language Models

arXiv:2606. 26079v1 Announce Type: cross Abstract: Standard benchmarks for multimodal large language models (MLLMs) score each item on one canonical ordering and miss whether order-irrelevant shuffling changes the answer, a baseline reliability property called for by emerging AI evaluation guidelines.

By Akshay Paruchuri, Sanmi Koyejo, Ehsan Adeli
llmsmultimodalbenchmarkssafety
More like this →
arXiv Computation and Language
Sep 25

Likelihood Ranking doesn't Scale Like Prompting in LLMs

The paper compares two common ways of evaluating large language models (LLMs): prompting them to answer questions directly and scoring candidate answers using likelihood-based metrics. The authors introduce a new protocol that ranks declarative statements derived from question–answer pairs, and test it across 95 decoder-only models (0.1B–104B parameters) on 10 multiple-choice QA datasets. They find that while prompted answering accuracy improves sharply with model scale and instruction tuning, statement‑likelihood ranking accuracy stays relatively stable, indicating that the two evaluation methods probe different aspects of model behavior.

By Alessandro Bondielli, Lucia Passaro, Davide Bacciu, Alessandro Lenci
llms
More like this →
Hugging Face Trending Papers
Jun 24

Same Evidence, Different Answer: Auditing Order Sensitivity in Multimodal Large Language Models

Standard benchmarks for multimodal large language models (MLLMs) score each item on one canonical ordering and miss whether order-irrelevant shuffling changes the answer, a baseline reliability property called for by emerging AI evaluation guidelines. We introduce Facet-Probe, a five-facet audit (option, evidence-chunk, document-rank, image-set, and mixed-modality ordering) of 18 frontier and open-weight MLLMs.

llmsmultimodalbenchmarkssafety
More like this →