Summary of Gsm-symbolic: Understanding the Limitations Of Mathematical Reasoning in Large Language Models, by Iman Mirzadeh et al.
GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models
by Iman Mirzadeh, Keivan Alizadeh, Hooman Shahrokhi, Oncel Tuzel, Samy Bengio, Mehrdad Farajtabar
First submitted to arxiv on: 7 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- 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 The paper investigates the formal reasoning capabilities of Large Language Models (LLMs) on grade-school-level mathematical questions, specifically on the GSM8K benchmark. While LLMs have shown significant improvement on this benchmark, it is unclear whether their reasoning capabilities have genuinely advanced. The authors introduce GSM-Symbolic, an improved benchmark that generates diverse questions using symbolic templates. This allows for more controlled evaluations and provides reliable metrics to measure the reasoning capabilities of LLMs. The study reveals that LLMs exhibit noticeable variance in responding to different instantiations of the same question, and their performance declines when only numerical values are altered. Additionally, the paper shows that LLMs’ performance deteriorates as the number of clauses in a question increases, suggesting that they replicate reasoning steps from their training data rather than performing genuine logical reasoning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large Language Models (LLMs) have gotten really good at math problems, but it’s unclear if they’re actually understanding what they’re doing. To figure out how well they’re doing, researchers created a special test called GSM-Symbolic that asks different kinds of math questions. They found that LLMs get confused when the numbers in the question change, and their answers get worse as the question gets longer. This means that LLMs are probably just copying what they learned from their training data instead of actually doing the math themselves. |
