Does Representation Intervention Really Identify Desired Concepts and Elicit Alignment?

• URL: http://arxiv.org/abs/2505.18672v1
• Date: Sat, 24 May 2025 12:23:52 GMT
• Title: Does Representation Intervention Really Identify Desired Concepts and Elicit Alignment?
• Authors: Hongzheng Yang, Yongqiang Chen, Zeyu Qin, Tongliang Liu, Chaowei Xiao, Kun Zhang, Bo Han,
• Abstract summary: Representation intervention aims to locate and modify the representations that encode the underlying concepts in Large Language Models.<n>If the interventions are faithful, the intervened LLMs should erase the harmful concepts and be robust to both in-distribution adversarial prompts and the out-of-distribution jailbreaks.<n>We propose Concept Concentration (COCA), which simplifies the decision boundary between harmful and benign representations.
• Score: 73.80382983108997
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Representation intervention aims to locate and modify the representations that encode the underlying concepts in Large Language Models (LLMs) to elicit the aligned and expected behaviors. Despite the empirical success, it has never been examined whether one could locate the faithful concepts for intervention. In this work, we explore the question in safety alignment. If the interventions are faithful, the intervened LLMs should erase the harmful concepts and be robust to both in-distribution adversarial prompts and the out-of-distribution (OOD) jailbreaks. While it is feasible to erase harmful concepts without degrading the benign functionalities of LLMs in linear settings, we show that it is infeasible in the general non-linear setting. To tackle the issue, we propose Concept Concentration (COCA). Instead of identifying the faithful locations to intervene, COCA refractors the training data with an explicit reasoning process, which firstly identifies the potential unsafe concepts and then decides the responses. Essentially, COCA simplifies the decision boundary between harmful and benign representations, enabling more effective linear erasure. Extensive experiments with multiple representation intervention methods and model architectures demonstrate that COCA significantly reduces both in-distribution and OOD jailbreak success rates, and meanwhile maintaining strong performance on regular tasks such as math and code generation.

Related papers

• Bridging the Gap in Vision Language Models in Identifying Unsafe Concepts Across Modalities [23.165174248333212]
  Vision-language models (VLMs) are increasingly applied to identify unsafe or inappropriate images.<n>It is still unclear whether they can recognize various unsafe concepts when presented in different modalities, such as text and images.<n>We conduct a systematic evaluation of VLMs' perception (concept recognition) and alignment (ethical reasoning) capabilities.<n>We introduce a simplified reinforcement learning (RL)-based approach using proximal policy optimization (PPO) to strengthen the ability to identify unsafe concepts from images.
  arXiv  Detail & Related papers  (2025-07-15T10:04:27Z)
• Probing the Robustness of Large Language Models Safety to Latent Perturbations [30.16804362984161]
  Safety alignment is a key requirement for building reliable Artificial General Intelligence.<n>We observe that minor latent shifts can still trigger unsafe responses in aligned models.<n>We introduce Layer-wise Adversarial Patch Training(LAPT), a fine-tuning strategy that injects controlled perturbations into hidden representations during training.
  arXiv  Detail & Related papers  (2025-06-19T07:03:05Z)
• Concept-Level Explainability for Auditing & Steering LLM Responses [12.089513278445704]
  ConceptX is a model-agnostic, concept-level explainability method.<n>It identifies semantically rich tokens in the prompt, and assigns them importance based on the outputs' semantic similarity.<n>It enables both auditing, by uncovering sources of bias, and steering, by modifying prompts to shift the sentiment or reduce the harmfulness of LLM responses.
  arXiv  Detail & Related papers  (2025-05-12T14:31:51Z)
• Cannot See the Forest for the Trees: Invoking Heuristics and Biases to Elicit Irrational Choices of LLMs [83.11815479874447]
  We propose a novel jailbreak attack framework, inspired by cognitive decomposition and biases in human cognition.<n>We employ cognitive decomposition to reduce the complexity of malicious prompts and relevance bias to reorganize prompts.<n>We also introduce a ranking-based harmfulness evaluation metric that surpasses the traditional binary success-or-failure paradigm.
  arXiv  Detail & Related papers  (2025-05-03T05:28:11Z)
• Improving LLM Safety Alignment with Dual-Objective Optimization [65.41451412400609]
  Existing training-time safety alignment techniques for large language models (LLMs) remain vulnerable to jailbreak attacks.<n>We propose an improved safety alignment that disentangles DPO objectives into two components: (1) robust refusal training, which encourages refusal even when partial unsafe generations are produced, and (2) targeted unlearning of harmful knowledge.
  arXiv  Detail & Related papers  (2025-03-05T18:01:05Z)
• Root Defence Strategies: Ensuring Safety of LLM at the Decoding Level [10.476222570886483]
  Large language models (LLMs) have demonstrated immense utility across various industries.<n>As LLMs advance, the risk of harmful outputs increases due to incorrect or malicious instruction prompts.<n>This paper examines the LLMs' capability to recognize harmful outputs, revealing and quantifying their proficiency in assessing the danger of previous tokens.
  arXiv  Detail & Related papers  (2024-10-09T12:09:30Z)
• PaCE: Parsimonious Concept Engineering for Large Language Models [57.740055563035256]
  Parsimonious Concept Engineering (PaCE) is a novel activation engineering framework for alignment. We construct a large-scale concept dictionary in the activation space, in which each atom corresponds to a semantic concept. We show that PaCE achieves state-of-the-art alignment performance while maintaining linguistic capabilities.
  arXiv  Detail & Related papers  (2024-06-06T17:59:10Z)
• DeAL: Decoding-time Alignment for Large Language Models [59.63643988872571]
  Large Language Models (LLMs) are nowadays expected to generate content aligned with human preferences. We propose DeAL, a framework that allows the user to customize reward functions and enables Detime Alignment of LLMs. Our experiments show that we can DeAL with fine-grained trade-offs, improve adherence to alignment objectives, and address residual gaps in LLMs.
  arXiv  Detail & Related papers  (2024-02-05T06:12:29Z)
• Fundamental Limitations of Alignment in Large Language Models [16.393916864600193]
  An important aspect in developing language models that interact with humans is aligning their behavior to be useful and unharmful. This is usually achieved by tuning the model in a way that enhances desired behaviors and inhibits undesired ones. We propose a theoretical approach called Behavior Expectation Bounds (BEB) which allows us to formally investigate several inherent characteristics and limitations of alignment in large language models.
  arXiv  Detail & Related papers  (2023-04-19T17:50:09Z)
• Exploring Robustness of Unsupervised Domain Adaptation in Semantic Segmentation [74.05906222376608]
  We propose adversarial self-supervision UDA (or ASSUDA) that maximizes the agreement between clean images and their adversarial examples by a contrastive loss in the output space. This paper is rooted in two observations: (i) the robustness of UDA methods in semantic segmentation remains unexplored, which pose a security concern in this field; and (ii) although commonly used self-supervision (e.g., rotation and jigsaw) benefits image tasks such as classification and recognition, they fail to provide the critical supervision signals that could learn discriminative representation for segmentation tasks.
  arXiv  Detail & Related papers  (2021-05-23T01:50:44Z)

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.