Summary of How Likely Do Llms with Cot Mimic Human Reasoning?, by Guangsheng Bao et al.
How Likely Do LLMs with CoT Mimic Human Reasoning?
by Guangsheng Bao, Hongbo Zhang, Cunxiang Wang, Linyi Yang, Yue Zhang
First submitted to arxiv on: 25 Feb 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 This paper investigates Chain-of-thought (CoT), a technique for Large Language Models (LLMs) to reason like humans. While CoT shows promise, its effectiveness is not always guaranteed, leaving questions about its usage unanswered. To diagnose the underlying mechanism, this study compares human and LLM reasoning processes using causal analysis. The results reveal that LLMs often deviate from the ideal causal chain, leading to spurious correlations and potential consistency errors. Factors such as in-context learning with examples strengthen the causal structure, whereas post-training techniques like supervised fine-tuning and reinforcement learning on human feedback weaken it. Surprisingly, increasing model size alone cannot improve the causal structure, highlighting the need for new research. By understanding how LLMs reason, this study aims to shed light on improving their reasoning capabilities. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how Large Language Models (LLMs) think and reason like humans. Currently, a technique called Chain-of-thought is not always reliable in making LLMs better reasoners. To figure out what’s going wrong, the researchers compared how humans and LLMs think about problems. They found that LLMs often take shortcuts or make mistakes when trying to solve problems. The study also looked at what makes LLMs more likely to follow a good thinking pattern. Surprisingly, just making the model bigger isn’t enough – new approaches are needed. |
Keywords
* Artificial intelligence * Fine tuning * Reinforcement learning * Supervised
