Not Search, But Scan: Benchmarking MLLMs on Scan-Oriented Academic Paper Reasoning

1. 1Identify the limitations of current search-oriented academic paper reasoning approaches
2. 2Develop scan-oriented benchmarks to evaluate MLLMs' ability to reason without pre-specified targets
3. 3Evaluate MLLMs on these benchmarks to identify areas for improvement
4. 4Use the results to fine-tune and improve MLLMs' reasoning capabilities

AI researchers and engineers working on large language models can benefit from this benchmark to improve their models' reasoning capabilities, and researchers in academia can use this to evaluate the effectiveness of MLLMs in assisting with research tasks

💡 Current MLLMs are limited by their reliance on search-oriented paradigms and need to be evaluated on scan-oriented tasks to achieve autonomous research capabilities

🚀 Benchmarking MLLMs on scan-oriented academic paper reasoning to advance autonomous research!