# Focus on Your Question! Interpreting and Mitigating Toxic CoT Problems in Commonsense Reasoning

arXiv: https://arxiv.org/abs/2402.18344 · HTML: https://mlsanity.com/paper/2402.18344/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning.
- Builds on 12 of its 44 references that are in the map.
- First posted on arXiv 28 Feb 2024.

## Placed in

- LLM reasoning / [Prompting and chain-of-thought reasoning](https://mlsanity.com/lr/prompting-and-chain-of-thought-reasoning/)

## Builds on (most cited first, 12 of 12)

- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Lost in the Middle: How Language Models Use Long Contexts](https://mlsanity.com/paper/2307.03172/) (2023)
- [Self-Refine: Iterative Refinement with Self-Feedback](https://mlsanity.com/paper/2303.17651/) (2023)
- [Measuring Faithfulness in Chain-of-Thought Reasoning](https://mlsanity.com/paper/2307.13702/) (2023)
- [Baichuan 2: Open Large-scale Language Models](https://mlsanity.com/paper/2309.10305/) (2023)
- [Faithful Chain-of-Thought Reasoning](https://mlsanity.com/paper/2301.13379/) (2023)
- [FOLIO: Natural Language Reasoning with First-Order Logic](https://mlsanity.com/paper/2209.00840/) (2022)
- [Rethinking with Retrieval: Faithful Large Language Model Inference](https://mlsanity.com/paper/2301.00303/) (2022)
- [Contrastive Chain-of-Thought Prompting](https://mlsanity.com/paper/2311.09277/) (2023)
- [Tailoring Self-Rationalizers with Multi-Reward Distillation](https://mlsanity.com/paper/2311.02805/) (2023)
- [Self-Contradictory Reasoning Evaluation and Detection](https://mlsanity.com/paper/2311.09603/) (2023)

## Built on by (most cited first, 11 of 11)

- [LLM-based Agents Suffer from Hallucinations: A Survey of Taxonomy, Methods, and Directions](https://mlsanity.com/paper/2509.18970/) (2025)
- [Instance-adaptive Zero-shot Chain-of-Thought Prompting](https://mlsanity.com/paper/2409.20441/) (2024)
- [OCEAN: Offline Chain-of-thought Evaluation and Alignment in Large Language Models](https://mlsanity.com/paper/2410.23703/) (2024)
- [Don't Take Things Out of Context: Attention Intervention for Enhancing Chain-of-Thought Reasoning in Large Language Models](https://mlsanity.com/paper/2503.11154/) (2025)
- [LLaVA-RadZ: Can Multimodal Large Language Models Effectively Tackle Zero-shot Radiology Recognition?](https://mlsanity.com/paper/2503.07487/) (2025)
- [LINKED: Eliciting, Filtering and Integrating Knowledge in Large Language Model for Commonsense Reasoning](https://mlsanity.com/paper/2410.09541/) (2024)
- [Unveiling and Addressing Pseudo Forgetting in Large Language Models](https://mlsanity.com/paper/2411.11932/) (2024)
- [Enhancing Chain-of-Thought Reasoning with Critical Representation Fine-tuning](https://mlsanity.com/paper/2507.10085/) (2025)
- [Toxicity in Online Platforms and AI Systems: A Survey of Needs, Challenges, Mitigations, and Future Directions](https://mlsanity.com/paper/2509.25539/) (2025)
- [OncoReason: Structuring Clinical Reasoning in LLMs for Robust and Interpretable Survival Prediction](https://mlsanity.com/paper/2510.17532/) (2025)
- [SalaMAnder: Shapley-based Mathematical Expression Attribution and Metric for Chain-of-Thought Reasoning](https://mlsanity.com/paper/2509.16561/) (2025)

## Abstract (arXiv)

Large language models exhibit high-level commonsense reasoning abilities, especially with enhancement methods like Chain-of-Thought (CoT). However, we find these CoT-like methods lead to a considerable number of originally correct answers turning wrong, which we define as the Toxic CoT problem. To interpret and mitigate this problem, we first utilize attribution tracing and causal tracing methods to probe the internal working mechanism of the LLM during CoT reasoning. Through comparisons, we prove that the model exhibits information loss from the question over the shallow attention layers when generating rationales or answers. Based on the probing findings, we design a novel method called RIDERS (Residual decodIng and sERial-position Swap), which compensates for the information deficit in the model from both decoding and serial-position perspectives. Through extensive experiments on multiple commonsense reasoning benchmarks, we validate that this method not only significantly eliminates Toxic CoT problems (decreased by 23.6%), but also effectively improves the model's overall commonsense reasoning performance (increased by 5.5%).