AI Paper Atlas

Research map / Large language models / Agents, code and math

M3Exam: A Multilingual, Multimodal, Multilevel Benchmark for Examining Large Language Models

Wenxuan Zhang, Mahani Aljunied, Chang Gao, Yew Ken Chia, Lidong Bing

NeurIPS 2023 · Large language models › Agents, code and math · approach: Long-horizon LLM agents · 32 citations (OpenAlex)

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 changedWhereMeasuredEffect
EN-Translation prompt instead of monolingual (ablation)ChatGPT · M3Exam Thaiaccuracy: 34.09 -> 46.49Improved
EN-Translation prompt instead of monolingual (ablation)ChatGPT · M3Exam Javaneseaccuracy: 37.47 -> 50.94Improved
EN-Translation prompt instead of monolingual (ablation)ChatGPT · M3Exam Italianaccuracy: 67.94 -> 62.76Hurt
EN-Instruct prompt instead of monolingual (ablation)ChatGPT · M3Exam (non-English)accuracy: Thai 34.09 -> 32.70; Italian 67.94 -> 69.30No clear effect
add few-shot demonstrations (ablation)ChatGPT · M3Exam (non-English)accuracy: Vietnamese 57.18 -> 58.64; Javanese 37.47 -> 33.42Mixed
GPT-4 vs ChatGPTGPT-4 · M3Exam avgaccuracy: 57.57 -> 72.92Improved
BLIP-2 vision input vs text-only Flan-T5BLIP-2 · M3Exam English image questionsaccuracy: 48.30 -> 49.06No clear effect
multi-image pretrained models (Fromage, OpenFlamingo) vs single-image BLIP-2multimodal LLMs · M3Exam English image questionsaccuracy: 49.06 -> 29.81Hurt
increase education level (low to high)LLMs · M3ExamaccuracyNo clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy