Research map / Vision-language and multimodal / Visual question answering
Choices Speak Louder than Questions
Gyeongje Cho, Yeonkyoung So, Jaejin Lee
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace log-likelihood scoring with NPSQ under simple adversarial choice | Llama-3.1-8B-Instruct · HellaSwag | predictions switched to adversarial: 93.19% -> under 0.17% | Improved |
| replace length-normalized scoring with NPSQ under extended adversarial choice | Llama-3.1-8B-Instruct · ARC-Challenge | accuracy drop: 18.17% drop -> below 0.05% difference | Improved |
| replace acc with NPSQ under instructional adversarial choice | Llama-3.1-8B-Instruct · MMLU | predictions switched: 27.47% -> 10.13% | Improved |
| replace acc with NPSQ under neutral adversarial choice | Llama-3.1-8B-Instruct · MMLU | accuracy change: -8.60% (acc) -> +3.31% (NPSQ) | Improved |
| use NPSQ instead of acc in cloze format | Llama-3.1-8B-Instruct · MMLU | reported accuracy: higher score than acc and acc_norm | Improved |
| use NPSQ instead of acc in symbols and hybrid formats | Llama-3.1-8B-Instruct · MMLU | reported accuracy: slightly lower scores | Hurt |
| use NPSQ for ranking models | various LLMs · MCQA benchmarks | model ranking: rankings reorder under NPSQ | Mixed |
Closest papers by meaning
- Answer, Assemble, Ace: Understanding How LMs Answer Multiple Choice Questions (ICLR 2025 · similarity 0.89)
- Rethinking Generative Large Language Model Evaluation for Semantic Comprehension (ICML 2024 · similarity 0.89)
- Evaluating language models as risk scores (NeurIPS 2024 · similarity 0.88)
- Large Language Models Are Not Robust Multiple Choice Selectors (ICLR 2024 · similarity 0.88)
- A Semantic-Sampling Framework for Evaluating Calibration in Open-Ended Question Answering (NeurIPS 2026 · similarity 0.86)
- Fool Your (Vision and) Language Model with Embarrassingly Simple Permutations (ICML 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
