Research map / Large language models / Agents, code and math
A Causality-Aware Paradigm for Evaluating Creativity of Multimodal Large Language Models
Zeyu Huang, Shanshan Zhong, Pan Zhou, Shanghua Gao, Marinka Žitnik, Liang Lin
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 |
|---|---|---|---|
| add associable instruction tuning (AITv2) (ablation) | Qwen-VL · Oogiri-GO IT2T | average accuracy (%): 33.6 -> 41.0 | Improved |
| add associable instruction tuning (AITv2) (ablation) | Qwen-VL · Oogiri-GO I2T | average accuracy (%): 30.1 -> 37.7 | Improved |
| add associable instruction tuning (AITv2) (ablation) | Qwen-VL · Oogiri-GO T2T | average accuracy (%): 30.3 -> 35.6 | Improved |
| add explorative self-refinement on top of AITv2 (full CLoTv2) (ablation) | Qwen-VL · Oogiri-GO IT2T | average accuracy (%): 41.0 -> 43.2 | Improved |
| add explorative self-refinement on top of AITv2 (full CLoTv2) (ablation) | Qwen-VL · Oogiri-GO I2T | average accuracy (%): 37.7 -> 41.8 | Improved |
| add explorative self-refinement on top of AITv2 (full CLoTv2) (ablation) | Qwen-VL · Oogiri-GO T2T | average accuracy (%): 35.6 -> 39.4 | Improved |
| add associable instruction tuning (AITv2) (ablation) | CogVLM-17B · Oogiri-GO I2T (English) | average accuracy (%): 29.9 -> 44.6 | Improved |
| add CLoTv2 (AIT plus self-refinement) | CogVLM-17B · Oogiri-GO I2T (English) | average accuracy (%): 29.9 -> 53.3 | Improved |
| add CLoTv2 (AIT plus self-refinement) | CogVLM-17B · Oogiri-GO T2T (English) | average accuracy (%): 31.9 -> 49.8 | Improved |
| replace CLoTv1 with CLoTv2 | Qwen-VL · Oogiri-GO I2T | average accuracy (%): 40.5 -> 41.8 | Improved |
| replace CLoTv1 with CLoTv2 | CogVLM-17B · Oogiri-GO I2T (English) | average accuracy (%): 51.6 -> 53.3 | Improved |
| use CoT-style prompting (CoT, CoT-SC, DDCoT) instead of CLoT training (ablation) | Qwen-VL · Oogiri-GO I2T | accuracy | No clear effect |
Closest papers by meaning
- Let's Think Outside the Box: Exploring Leap-of-Thought in Large Language Models with Creative Humor Generation (CVPR 2024 · similarity 0.92)
- CreBench: Human-Aligned Creativity Evaluation from Idea to Process to Product (AAAI 2026 · similarity 0.90)
- IDEAFix: Evaluation Framework for Creative Defixation Prompting in LLMs (NeurIPS 2026 · similarity 0.89)
- CLAWS:Creativity detection for LLM-generated solutions using Attention Window of Sections (NeurIPS 2025 · similarity 0.89)
- When Visualizing is the First Step to Reasoning: MIRA, a Benchmark for Visual Chain-of-Thought (CVPR 2026 · similarity 0.88)
- AI as Humanity’s Salieri: Quantifying Linguistic Creativity of Language Models via Systematic Attribution of Machine Text against Web Text (ICLR 2025 · similarity 0.88)
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.
