Summary of Focus on Your Question! Interpreting and Mitigating Toxic Cot Problems in Commonsense Reasoning, by Jiachun Li et al.
Focus on Your Question! Interpreting and Mitigating Toxic CoT Problems in Commonsense Reasoning
by Jiachun Li, Pengfei Cao, Chenhao Wang, Zhuoran Jin, Yubo Chen, Daojian Zeng, Kang Liu, Jun Zhao
First submitted to arxiv on: 28 Feb 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 The paper investigates the limitations of Chain-of-Thought (CoT) methods in large language models, which can lead to incorrect answers despite initial correctness. The authors use attribution tracing and causal tracing techniques to understand the model’s internal workings during CoT reasoning. They find that information loss occurs due to shallow attention layers. To address this issue, they propose a novel method called RIDERS, which improves decoding and serial-position processing. This approach significantly reduces Toxic CoT problems (23.6%) and boosts overall commonsense reasoning performance (5.5%). |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can think deeply about questions. However, some methods make them more likely to give wrong answers even if they were correct at first. The authors want to understand why this happens. They use special tools to see how the model works during this process and find that it loses information along the way. To fix this problem, they developed a new method called RIDERS. This helps the model remember more details and give better answers. |
Keywords
» Artificial intelligence » Attention
