Summary of Mm-phyqa: Multimodal Physics Question-answering with Multi-image Cot Prompting, by Avinash Anand et al.
MM-PhyQA: Multimodal Physics Question-Answering With Multi-Image CoT Prompting
by Avinash Anand, Janak Kapuriya, Apoorv Singh, Jay Saraf, Naman Lal, Astha Verma, Rushali Gupta, Rajiv Shah
First submitted to arxiv on: 11 Apr 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper investigates the limitations of Large Language Models (LLMs) in tackling multi-step physics reasoning tasks. To address this challenge, the authors created a novel dataset called MM-PhyQA, comprising high school-level multimodal physics problems. They evaluated the performance of contemporary LLMs, including GPT-4 and various models from the LLaVA family, using zero-shot prediction and fine-tuning on their dataset. The study also explored the effectiveness of the Multi-Image Chain-of-Thought (MI-CoT) Prompting technique in improving model performance. The results show that the best-performing model, LLaVA-1.5 13b, achieved an accuracy of 71.65% on the test set using MI-CoT prompting. This research aims to identify the shortcomings of existing models and facilitate further development in this area. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how well computers can understand complex physics problems that involve multiple steps. The authors created a new dataset with many different types of physics questions that include images and text. They tested some powerful language models on these questions to see how well they could do. One model, called LLaVA-1.5 13b, did very well when it was trained using a special technique. This research helps us understand what computers are good at and where we need to improve. |
Keywords
» Artificial intelligence » Fine tuning » Gpt » Prompting » Zero shot
