Aligned Probing: Relating Toxic Behavior and Model Internals

• URL: http://arxiv.org/abs/2503.13390v1
• Date: Mon, 17 Mar 2025 17:23:50 GMT
• Title: Aligned Probing: Relating Toxic Behavior and Model Internals
• Authors: Andreas Waldis, Vagrant Gautam, Anne Lauscher, Dietrich Klakow, Iryna Gurevych,
• Abstract summary: We introduce aligned probing, a novel interpretability framework that aligns the behavior of language models (LMs)<n>Using this framework, we examine over 20 OLMo, Llama, and Mistral models, bridging behavioral and internal perspectives for toxicity for the first time.<n>Our results show that LMs strongly encode information about the toxicity level of inputs and subsequent outputs, particularly in lower layers.
• Score: 66.49887503194101
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: We introduce aligned probing, a novel interpretability framework that aligns the behavior of language models (LMs), based on their outputs, and their internal representations (internals). Using this framework, we examine over 20 OLMo, Llama, and Mistral models, bridging behavioral and internal perspectives for toxicity for the first time. Our results show that LMs strongly encode information about the toxicity level of inputs and subsequent outputs, particularly in lower layers. Focusing on how unique LMs differ offers both correlative and causal evidence that they generate less toxic output when strongly encoding information about the input toxicity. We also highlight the heterogeneity of toxicity, as model behavior and internals vary across unique attributes such as Threat. Finally, four case studies analyzing detoxification, multi-prompt evaluations, model quantization, and pre-training dynamics underline the practical impact of aligned probing with further concrete insights. Our findings contribute to a more holistic understanding of LMs, both within and beyond the context of toxicity.

Related papers

• Large Language Models can be Strong Self-Detoxifiers [82.6594169242814]
  Self-disciplined Autoregressive Sampling (SASA) is a lightweight controlled decoding algorithm for toxicity reduction of large language models (LLMs) SASA tracks the margin of the current output to steer the generation away from the toxic subspace, by adjusting the autoregressive sampling strategy. evaluated on LLMs of different scale and nature, namely Llama-3.1-Instruct (8B), Llama-2 (7B), and GPT2-L models with the RealToxicityPrompts, BOLD, and AttaQ benchmarks.
  arXiv  Detail & Related papers  (2024-10-04T17:45:15Z)
• Detoxifying Large Language Models via Knowledge Editing [57.0669577257301]
  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. We conduct experiments with several knowledge editing approaches, indicating that knowledge editing has the potential to detoxify LLMs with a limited impact on general performance efficiently.
  arXiv  Detail & Related papers  (2024-03-21T15:18:30Z)
• CausalGym: Benchmarking causal interpretability methods on linguistic tasks [52.61917615039112]
  We use CausalGym to benchmark the ability of interpretability methods to causally affect model behaviour. We study the pythia models (14M--6.9B) and assess the causal efficacy of a wide range of interpretability methods. We find that DAS outperforms the other methods, and so we use it to study the learning trajectory of two difficult linguistic phenomena.
  arXiv  Detail & Related papers  (2024-02-19T21:35:56Z)
• Retrieve Only When It Needs: Adaptive Retrieval Augmentation for Hallucination Mitigation in Large Language Models [68.91592125175787]
  Hallucinations pose a significant challenge for the practical implementation of large language models (LLMs) We present Rowen, a novel approach that enhances LLMs with a selective retrieval augmentation process tailored to address hallucinations.
  arXiv  Detail & Related papers  (2024-02-16T11:55:40Z)
• Can LLMs Recognize Toxicity? A Structured Investigation Framework and Toxicity Metric [16.423707276483178]
  We introduce a robust metric grounded on Large Language Models (LLMs) to flexibly measure toxicity according to the given definition. Our results demonstrate outstanding performance in measuring toxicity within verified factors, improving on conventional metrics by 12 points in the F1 score.
  arXiv  Detail & Related papers  (2024-02-10T07:55:27Z)
• Unveiling the Implicit Toxicity in Large Language Models [77.90933074675543]
  The open-endedness of large language models (LLMs) combined with their impressive capabilities may lead to new safety issues when being exploited for malicious use. We show that LLMs can generate diverse implicit toxic outputs that are exceptionally difficult to detect via simply zero-shot prompting. We propose a reinforcement learning (RL) based attacking method to further induce the implicit toxicity in LLMs.
  arXiv  Detail & Related papers  (2023-11-29T06:42:36Z)
• Toxicity Detection with Generative Prompt-based Inference [3.9741109244650823]
  It is a long-known risk that language models (LMs), once trained on corpus containing undesirable content, have the power to manifest biases and toxicity. In this work, we explore the generative variant of zero-shot prompt-based toxicity detection with comprehensive trials on prompt engineering.
  arXiv  Detail & Related papers  (2022-05-24T22:44:43Z)
• Leashing the Inner Demons: Self-Detoxification for Language Models [13.576289320208511]
  Language models (LMs) can reproduce (or amplify) toxic language seen during training. We analyze the impact of prompts, decoding strategies and training corpora on the output. We propose a simple yet effective method for language models to "detoxify" themselves without an additional large corpus or external discriminator.
  arXiv  Detail & Related papers  (2022-03-06T23:55:12Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.