Summary of Mmlu-sr: a Benchmark For Stress-testing Reasoning Capability Of Large Language Models, by Wentian Wang et al.
MMLU-SR: A Benchmark for Stress-Testing Reasoning Capability of Large Language Models
by Wentian Wang, Sarthak Jain, Paul Kantor, Jacob Feldman, Lazaros Gallos, Hao Wang
First submitted to arxiv on: 15 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 A novel dataset, MMLU-SR, is proposed to measure the true comprehension abilities of Large Language Models (LLMs) by challenging their performance in question-answering tasks with modified terms. The approach replaces key terms with defined alternate terms to differentiate between comprehension and mere text replacement. In a study, standardized test questions were modified by replacing a key term with a dummy word along with its definition, both in the context of questions, answers, or both. Despite high scores achieved by popular LLMs on the MMLU leaderboard, model performance was found to be substantially reduced after such replacement, suggesting poor comprehension. |
Low | GrooveSquid.com (original content) | Low Difficulty Summary A new way to test how well computers understand language is created. The method changes important words in questions and answers to see if computer models can still answer correctly. This helps us know if a model truly understands something or just copies the words. Some popular computer models are good at answering questions, but they do poorly when the key words are changed. This new test will help scientists make better computer models that can really understand language. |
Keywords
» Artificial intelligence » Question answering