Daily Papers

This paper investigates using knowledge editing techniques to detoxify Large Language Models (LLMs). We construct a benchmark, SafeEdit, which covers nine unsafe categories with various powerful attack prompts and equips comprehensive metrics for systematic evaluation. We conduct experiments to compare knowledge editing approaches with previous baselines, indicating that knowledge editing has the potential to efficiently detoxify LLMs with limited impact on general performance. Then, we propose a simple yet effective baseline, dubbed Detoxifying with Intraoperative Neural Monitoring (DINM), to diminish the toxicity of LLMs within a few tuning steps via only one instance. We further provide an in-depth analysis of the internal mechanism for various detoxify approaches, demonstrating that previous methods like SFT and DPO may merely suppress the activations of toxic parameters, while DINM mitigates the toxicity of the toxic parameters to a certain extent, making permanent adjustments. We hope that these insights could shed light on future work of developing detoxifying approaches and the underlying knowledge mechanisms of LLMs. Code and benchmark are available at https://github.com/zjunlp/EasyEdit.

Text detoxification has the potential to mitigate the harms of toxicity by rephrasing text to remove offensive meaning, but subtle toxicity remains challenging to tackle. We introduce MaRCo, a detoxification algorithm that combines controllable generation and text rewriting methods using a Product of Experts with autoencoder language models (LMs). MaRCo uses likelihoods under a non-toxic LM (expert) and a toxic LM (anti-expert) to find candidate words to mask and potentially replace. We evaluate our method on several subtle toxicity and microaggressions datasets, and show that it not only outperforms baselines on automatic metrics, but MaRCo's rewrites are preferred 2.1 times more in human evaluation. Its applicability to instances of subtle toxicity is especially promising, demonstrating a path forward for addressing increasingly elusive online hate.

Language model detoxification aims to minimize the risk of generating offensive or harmful content in pretrained language models (PLMs) for safer deployment. Existing methods can be roughly categorized as finetuning-based and decoding-based. However, the former is often resource-intensive, while the latter relies on additional components and potentially compromises the generation fluency. In this paper, we propose a more lightweight approach that enables the PLM itself to achieve "self-detoxification". Our method is built upon the observation that prepending a negative steering prompt can effectively induce PLMs to generate toxic content. At the same time, we are inspired by the recent research in the interpretability field, which formulates the evolving contextualized representations within the PLM as an information stream facilitated by the attention layers. Drawing on this idea, we devise a method to identify the toxification direction from the normal generation process to the one prompted with the negative prefix, and then steer the generation to the reversed direction by manipulating the information movement within the attention layers. Experimental results show that our approach, without any fine-tuning or extra components, can achieve comparable performance with state-of-the-art methods.

Warning: this paper contains model outputs exhibiting offensiveness and biases. Recently pre-trained language models (PLMs) have prospered in various natural language generation (NLG) tasks due to their ability to generate fairly fluent text. Nevertheless, these models are observed to capture and reproduce harmful contents in training corpora, typically toxic language and social biases, raising severe moral issues. Prior works on ethical NLG tackle detoxifying and debiasing separately, which is problematic since we find debiased models still exhibit toxicity while detoxified ones even exacerbate biases. To address such a challenge, we propose the first unified framework of detoxifying and debiasing called UDDIA, which jointly formalizes these two problems as rectifying the output space. We theoretically interpret our framework as learning a text distribution mixing weighted attributes. Besides, UDDIA conducts adaptive optimization of only a few parameters during decoding based on a parameter-efficient tuning schema without any training data. This leads to minimal generation quality loss and improved rectification performance with acceptable computational cost. Experimental results demonstrate that compared to several strong baselines, UDDIA achieves debiasing and detoxifying simultaneously and better balances efficiency and effectiveness, taking a further step towards practical ethical NLG.

Large language models (LM) generate remarkably fluent text and can be efficiently adapted across NLP tasks. Measuring and guaranteeing the quality of generated text in terms of safety is imperative for deploying LMs in the real world; to this end, prior work often relies on automatic evaluation of LM toxicity. We critically discuss this approach, evaluate several toxicity mitigation strategies with respect to both automatic and human evaluation, and analyze consequences of toxicity mitigation in terms of model bias and LM quality. We demonstrate that while basic intervention strategies can effectively optimize previously established automatic metrics on the RealToxicityPrompts dataset, this comes at the cost of reduced LM coverage for both texts about, and dialects of, marginalized groups. Additionally, we find that human raters often disagree with high automatic toxicity scores after strong toxicity reduction interventions -- highlighting further the nuances involved in careful evaluation of LM toxicity.

Pre-trained language models (LMs) are shown to easily generate toxic language. In this work, we systematically explore domain-adaptive training to reduce the toxicity of language models. We conduct this study on three dimensions: training corpus, model size, and parameter efficiency. For the training corpus, we propose to leverage the generative power of LMs and generate nontoxic datasets for domain-adaptive training, which mitigates the exposure bias and is shown to be more data-efficient than using a curated pre-training corpus. We demonstrate that the self-generation method consistently outperforms the existing baselines across various model sizes on both automatic and human evaluations, even when it uses a 1/3 smaller training corpus. We then comprehensively study detoxifying LMs with parameter sizes ranging from 126M up to 530B (3x larger than GPT-3), a scale that has never been studied before. We find that i) large LMs have similar toxicity levels as smaller ones given the same pre-training corpus, and ii) large LMs require more endeavor to detoxify. We also explore parameter-efficient training methods for detoxification. We demonstrate that adding and training adapter-only layers in LMs not only saves a lot of parameters but also achieves a better trade-off between toxicity and perplexity than whole model adaptation for the large-scale models.

Language detoxification involves removing toxicity from offensive language. While a neutral-toxic paired dataset provides a straightforward approach for training detoxification models, creating such datasets presents several challenges: i) the need for human annotation to build paired data, and ii) the rapid evolution of offensive terms, rendering static datasets quickly outdated. To tackle these challenges, we introduce an automated paired data generation pipeline, called K/DA. This pipeline is designed to generate offensive language with implicit offensiveness and trend-aligned slang, making the resulting dataset suitable for detoxification model training. We demonstrate that the dataset generated by K/DA exhibits high pair consistency and greater implicit offensiveness compared to existing Korean datasets, and also demonstrates applicability to other languages. Furthermore, it enables effective training of a high-performing detoxification model with simple instruction fine-tuning.

Detoxifying multilingual Large Language Models (LLMs) has become crucial due to their increasing global use. In this work, we explore zero-shot cross-lingual generalization of preference tuning in detoxifying LLMs. Unlike previous studies that show limited cross-lingual generalization for other safety tasks, we demonstrate that Direct Preference Optimization (DPO) training with only English data can significantly reduce toxicity in multilingual open-ended generations. For example, the probability of mGPT-1.3B generating toxic continuations drops from 46.8% to 3.9% across 17 different languages after training. Our results also extend to other multilingual LLMs, such as BLOOM, Llama3, and Aya-23. Using mechanistic interpretability tools like causal intervention and activation analysis, we identified the dual multilinguality property of MLP layers in LLMs, which explains the cross-lingual generalization of DPO. Finally, we show that bilingual sentence retrieval can predict the cross-lingual transferability of DPO preference tuning.

Detoxifying offensive language while preserving the speaker's original intent is a challenging yet critical goal for improving the quality of online interactions. Although large language models (LLMs) show promise in rewriting toxic content, they often default to overly polite rewrites, distorting the emotional tone and communicative intent. This problem is especially acute in Chinese, where toxicity often arises implicitly through emojis, homophones, or discourse context. We present ToxiRewriteCN, the first Chinese detoxification dataset explicitly designed to preserve sentiment polarity. The dataset comprises 1,556 carefully annotated triplets, each containing a toxic sentence, a sentiment-aligned non-toxic rewrite, and labeled toxic spans. It covers five real-world scenarios: standard expressions, emoji-induced and homophonic toxicity, as well as single-turn and multi-turn dialogues. We evaluate 17 LLMs, including commercial and open-source models with variant architectures, across four dimensions: detoxification accuracy, fluency, content preservation, and sentiment polarity. Results show that while commercial and MoE models perform best overall, all models struggle to balance safety with emotional fidelity in more subtle or context-heavy settings such as emoji, homophone, and dialogue-based inputs. We release ToxiRewriteCN to support future research on controllable, sentiment-aware detoxification for Chinese.

Toxic content has become an increasingly critical social issue with the rapid expansion of online communication. While numerous studies explored methods for detecting and detoxifying such content, most have focused primarily on English, leaving low-resource language underrepresented. Consequently, Large Language Models~(LLMs) often struggle to identify and neutralize toxic expressions in these languages. This challenge becomes even more pronounced when user employ obfuscation techniques to evade detection systems. Therefore, we propose a KOTOX: Korean Toxic Dataset for deobfuscation and detoxicification to address this issue. We categorize various obfuscation approaches based on linguistic characteristics of Korean and define a set of transformation rules grounded in real-word examples. Using these rules, we construct three dataset versions (easy, normal, and hard) representing different levels of obfuscation difficulty. This is the first dataset that simultaneously supports deobfuscation and detoxification for the Korean language. We expect it to facilitate better understanding and mitigating of obfuscated toxic content in LLM for low-resource languages. Our code and data are available at https://github.com/leeyejin1231/KOTOX.

In large language model (LLM) pretraining, data quality is believed to determine model quality. In this paper, we re-examine the notion of "quality" from the perspective of pre- and post-training co-design. Specifically, we explore the possibility that pre-training on more toxic data can lead to better control in post-training, ultimately decreasing a model's output toxicity. First, we use a toy experiment to study how data composition affects the geometry of features in the representation space. Next, through controlled experiments with Olmo-1B models trained on varying ratios of clean and toxic data, we find that the concept of toxicity enjoys a less entangled linear representation as the proportion of toxic data increases. Furthermore, we show that although toxic data increases the generational toxicity of the base model, it also makes the toxicity easier to remove. Evaluations on Toxigen and Real Toxicity Prompts demonstrate that models trained on toxic data achieve a better trade-off between reducing generational toxicity and preserving general capabilities when detoxifying techniques such as inference-time intervention (ITI) are applied. Our findings suggest that, with post-training taken into account, bad data may lead to good models.

Text detoxification aims to minimize the risk of language models producing toxic content. Existing detoxification methods of directly constraining the model output or further training the model on the non-toxic corpus fail to achieve a decent balance between detoxification effectiveness and generation quality. This issue stems from the neglect of constrain imposed by the context since language models are designed to generate output that closely matches the context while detoxification methods endeavor to ensure the safety of the output even if it semantically deviates from the context. In view of this, we introduce a Context-aware Model self-Detoxification~(CMD) framework that pays attention to both the context and the detoxification process, i.e., first detoxifying the context and then making the language model generate along the safe context. Specifically, CMD framework involves two phases: utilizing language models to synthesize data and applying these data for training. We also introduce a toxic contrastive loss that encourages the model generation away from the negative toxic samples. Experiments on various LLMs have verified the effectiveness of our MSD framework, which can yield the best performance compared to baselines.

We propose a method to control the attributes of Language Models (LMs) for the text generation task using Causal Average Treatment Effect (ATE) scores and counterfactual augmentation. We explore this method, in the context of LM detoxification, and propose the Causally Fair Language (CFL) architecture for detoxifying pre-trained LMs in a plug-and-play manner. Our architecture is based on a Structural Causal Model (SCM) that is mathematically transparent and computationally efficient as compared with many existing detoxification techniques. We also propose several new metrics that aim to better understand the behaviour of LMs in the context of toxic text generation. Further, we achieve state of the art performance for toxic degeneration, which are computed using \RTP (RTP) benchmark. Our experiments show that CFL achieves such a detoxification without much impact on the model perplexity. We also show that CFL mitigates the unintended bias problem through experiments on the BOLD dataset.

While large-scale language models (LMs) are able to imitate the distribution of natural language well enough to generate realistic text, it is difficult to control which regions of the distribution they generate. This is especially problematic because datasets used for training large LMs usually contain significant toxicity, hate, bias, and negativity. We propose GeDi as an efficient method for using smaller LMs as generative discriminators to guide generation from large LMs to make them safer and more controllable. GeDi guides generation at each step by computing classification probabilities for all possible next tokens via Bayes rule by normalizing over two class-conditional distributions; one conditioned on the desired attribute, or control code, and another conditioned on the undesired attribute, or anti control code. We find that GeDi gives stronger controllability than the state of the art method while also achieving generation speeds more than 30 times faster. Additionally, training GeDi on only four topics allows us to controllably generate new topics zero-shot from just a keyword, unlocking a new capability that previous controllable generation methods do not have. Lastly, we show that GeDi can make GPT-2 (1.5B parameters) significantly less toxic without sacrificing linguistic quality, making it by far the most practical existing method for detoxifying large language models while maintaining a fast generation speed.