Pragmatic Inference Chain (PIC) Improving LLMs’ Reasoning of Authentic Implicit Toxic Language (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies have shown that LLMs can detect toxicity by using a variety of inference-intensive tasks, such as understanding humour and metaphors. |
| Approach: | They propose a new method to prompt LLMs to identify toxic language using a set of online data that are verified by human annotators. |
| Outcome: | The proposed method significantly improves the success rate of GPT-4o, Llama-3.1-70B-Instruct, DeepSeek-v2.5, and DeepSeq-v3 in identifying implicit toxic language compared to five baseline prompts, such as CoT and rule-based baselines. |
Similar Papers
Unveiling the Implicit Toxicity in Large Language Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies focus on probing toxic outputs that can be easily detected with existing toxicity classifiers, but LLMs can generate diverse implicit toxic output that are difficult to detect via simply zero-shot prompting. |
| Approach: | They propose a reinforcement learning based attacking method to induce the implicit toxic outputs in large language models by fine-tuning toxicity classifiers. |
| Outcome: | The proposed method generates implicit toxic outputs that are difficult to detect via zero-shot prompting on five widely-adopted toxicity classifiers. |
Realistic Evaluation of Toxicity in Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | a large amount of data exposes large language models to toxicity and bias . prompt engineering can be easily bypassed with minimal prompt engineering. |
| Approach: | They propose a dataset that uses manually crafted prompts to nullify protective layers of large language models. |
| Outcome: | The proposed dataset shows that prompts can nullify protective layers of large language models. |
♪ Something Just Like TRuST ♪ *: Toxicity Recognition of Span and Target (2026.findings-acl)
Copied to clipboard
| Challenge: | Toxic language is pervasive online, and because LLMs are trained on web data, it generates such content. |
| Approach: | They propose a large-scale dataset that synthesizes toxicity definitions and an annotation scheme . they use a rigorous human annotation process to evaluate the diversity of the annotations . |
| Outcome: | The proposed model outperforms existing models on three tasks and is not reliable. |
Walking in Others’ Shoes: How Perspective-Taking Guides Large Language Models in Reducing Toxicity and Bias (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing prompting methods that require white-box access to the model or substantial training fail to simultaneously lessen toxicity and bias. |
| Approach: | They propose a strategy that encourages LLMs to integrate diverse human perspectives and self-regulate their responses by incorporating diverse human viewpoints. |
| Outcome: | The proposed approach can significantly diminish toxicity (up to 89%) and bias (up 73%) in LLMs’ responses. |
Know Thy Enemy: Securing LLMs Against Prompt Injection via Diverse Data Synthesis and Instruction-Level Chain-of-Thought Learning (2026.findings-acl)
Copied to clipboard
Zhiyuan Chang, Mingyang Li, Yuekai Huang, Ziyou Jiang, Xiaojun Jia, Qian Xiong, Junjie Wang, Zhaoyang Li, Qing Wang
| Challenge: | Large language model (LLM)-integrated applications face security vulnerabilities from prompt injection (PI) attacks. |
| Approach: | They propose a model enhancement method that synthesizes diverse training data and employs instruction-level chain-of-thought fine-tuning to enable LLMs to effectively identify and reject malicious instructions regardless of their source or position in the context. |
| Outcome: | The proposed method outperforms baselines in three critical dimensions while maintaining utility performance without degradation. |
Probing LLMs for hate speech detection: strengths and vulnerabilities (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Recent efforts to detect hateful or toxic language using large language models have not used explanation, additional context and victim community information in the detection process. |
| Approach: | They use different prompt variations, input information and victim community information to evaluate large language models in zero shot setting without adding any in-context examples. |
| Outcome: | The proposed models perform significantly better when included in the pipeline than baseline models. |
Understanding and Mitigating Overrefusal in LLMs from an Unveiling Perspective of Safety Decision Boundary (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) often refuse to answer legitimate queries, causing models to treat many reasonable prompts as potentially risky. |
| Approach: | They propose a framework that automatically generates and selects overrefusal prompts near the safety boundary. |
| Outcome: | The proposed framework identifies and curates boundary-aligned prompts, enabling more effective and targeted mitigation of overrefusal. |
Don’t Go To Extremes: Revealing the Excessive Sensitivity and Calibration Limitations of LLMs in Implicit Hate Speech Detection (2024.acl-long)
Copied to clipboard
| Challenge: | Several studies have examined whether large language models exhibit bias or discrimination against individuals or groups in terms of protected attributes like race, gender, or religion. |
| Approach: | They evaluate LLMs' ability to detect implicit hate speech and express confidence in their responses by considering prompt patterns and mainstream uncertainty estimation methods. |
| Outcome: | The proposed models exhibit two extremes: (1) excessive sensitivity towards groups or topics that may cause fairness issues, resulting in misclassifying benign statements as hate speech; (2) confidence scores for each method excessively concentrate on a fixed range, remaining unchanged regardless of the dataset’s complexity. |
Can LLMs Recognize Toxicity? A Structured Investigation Framework and Toxicity Metric (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing toxicity metrics rely on encoder models trained on specific toxicity datasets, which are susceptible to out-of-distribution (OOD) problems and depend on the dataset’s definition of toxicity. |
| Approach: | They propose a robust metric grounded on LLMs to flexibly measure toxicity according to the given definition by analysing toxicity factors and intrinsic toxic attributes. |
| Outcome: | The proposed metric improves on conventional metrics by 12 points in the F1 score and shows that upstream toxicity significantly influences downstream metrics, suggesting that LLMs are unsuitable for toxicity evaluations within unverified factors. |
Enhancing LLM-based Hatred and Toxicity Detection with Meta-Toxic Knowledge Graph (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to address toxicity issues with large language models are inadequate . lack of domain-specific knowledge leads to false negatives and excessive sensitivity to toxic speech limits freedom of speech. |
| Approach: | They propose a method that leverages graph search on a meta-toxic knowledge graph to enhance hatred and toxicity detection. |
| Outcome: | The proposed method lowers false positive rate and improves toxicity detection performance in out-of-domain scenarios. |