English
Related papers

Related papers: Circuit Breaking: Removing Model Behaviors with Ta…

200 papers

Language models, while capable of generating remarkably coherent and seemingly accurate text, can occasionally produce undesirable content, including harmful or toxic outputs. In this paper, we present a new two-stage approach to detect and…

Machine Learning · Computer Science 2025-11-11 Bao Nguyen , Binh Nguyen , Duy Nguyen , Viet Anh Nguyen

Training a language model to be both helpful and harmless requires careful calibration of refusal behaviours: Models should refuse to follow malicious instructions or give harmful advice (e.g."how do I kill someone?"), but they should not…

Computation and Language · Computer Science 2025-03-05 Xinpeng Wang , Chengzhi Hu , Paul Röttger , Barbara Plank

Large Language Models (LLMs) are widely deployed in real-world applications, yet their internal mechanisms remain difficult to interpret and control, limiting our ability to diagnose and correct undesirable behaviors. Mechanistic…

Neural network models have achieved high performance on a wide variety of complex tasks, but the algorithms that they implement are notoriously difficult to interpret. It is often necessary to hypothesize intermediate variables involved in…

Computation and Language · Computer Science 2025-02-13 Michael A. Lepori , Thomas Serre , Ellie Pavlick

Mechanistic interpretability has identified small sets of attention heads that implement specific behaviours in transformer language models, but recovering these circuits typically requires a bespoke analytical pipeline for each new task.…

Machine Learning · Computer Science 2026-05-27 Barsat Khadka

Caution: This paper includes offensive words that could potentially cause unpleasantness. The fast-paced evolution of generative language models such as GPT-4 has demonstrated outstanding results in various NLP generation tasks. However,…

Computation and Language · Computer Science 2023-12-12 Heegyu Kim , Hyunsouk Cho

Language models often exhibit undesirable behavior, e.g., generating toxic or gender-biased text. In the case of neural language models, an encoding of the undesirable behavior is often present in the model's representations. Thus, one…

Machine Learning · Computer Science 2025-06-05 Shashwat Singh , Shauli Ravfogel , Jonathan Herzig , Roee Aharoni , Ryan Cotterell , Ponnurangam Kumaraguru

The path to interpreting a language model often proceeds via analysis of circuits -- sparse computational subgraphs of the model that capture specific aspects of its behavior. Recent work has automated the task of discovering circuits. Yet,…

Computation and Language · Computer Science 2025-04-03 Adithya Bhaskar , Alexander Wettig , Dan Friedman , Danqi Chen

Language models (LMs) exhibit and amplify many types of undesirable biases learned from the training data, including gender bias. However, we lack tools for effectively and efficiently changing this behavior without hurting general language…

Computation and Language · Computer Science 2023-10-20 Abhijith Chintam , Rahel Beloch , Willem Zuidema , Michael Hanna , Oskar van der Wal

We study attribute control in language models through the method of Causal Average Treatment Effect (Causal ATE). Existing methods for the attribute control task in Language Models (LMs) check for the co-occurrence of words in a sentence…

Computation and Language · Computer Science 2024-02-19 Rahul Madhavan , Kahini Wadhawan

Safety-aligned language models systematically refuse harmful requests. While activation steering can modulate refusal, ablating the raw "refusal vector" calculated from contrastive harmful and harmless prompts often causes collateral damage…

Computation and Language · Computer Science 2026-01-14 Tony Cristofano

Neural language models often fail to generate diverse and informative texts, limiting their applicability in real-world problems. While previous approaches have proposed to address these issues by identifying and penalizing undesirable…

Computation and Language · Computer Science 2023-09-25 Jimin Hong , ChaeHun Park , Jaegul Choo

How does language model pretraining help transfer learning? We consider a simple ablation technique for determining the impact of each pretrained layer on transfer task performance. This method, partial reinitialization, involves replacing…

Computation and Language · Computer Science 2020-11-11 Alex Tamkin , Trisha Singh , Davide Giovanardi , Noah Goodman

We present a mechanistic analysis of literary style in GPT-2, identifying individual neurons that discriminate between exemplary prose and rigid AI-generated text. Using Herman Melville's Bartleby, the Scrivener as a corpus, we extract…

Computation and Language · Computer Science 2025-10-22 Tsogt-Ochir Enkhbayar

We propose a five-stage methodology for causal feature analysis in transformer language models (probe design, feature extraction, causal validation, robustness testing, and deployment integration) and demonstrate it end-to-end on GPT-2…

Computation and Language · Computer Science 2026-05-22 Caleb Munigety

Deep generative models are known to produce undesirable samples such as harmful content. Traditional mitigation methods include re-training from scratch, filtering, or editing; however, these are either computationally expensive or can be…

Machine Learning · Computer Science 2024-02-22 Zhifeng Kong , Kamalika Chaudhuri

The circuits framework in mechanistic interpretability aims to identify causally important sparse subgraphs of model components, typically evaluated by measuring necessity and sufficiency. We measure circuit reuse, the proportion of…

Computation and Language · Computer Science 2026-05-12 Michael Li , Nishant Subramani

Large language models (LLMs) are now ubiquitous in user-facing applications, yet they still generate undesirable toxic outputs, including profanity, vulgarity, and derogatory remarks. Although numerous detoxification methods exist, most…

Computation and Language · Computer Science 2025-10-24 Agam Goyal , Vedant Rathi , William Yeh , Yian Wang , Yuen Chen , Hari Sundaram

Large language models (LLMs) are typically aligned to refuse harmful instructions through safety fine-tuning. A recent attack, termed abliteration, identifies and suppresses the single latent direction most responsible for refusal behavior,…

Computation and Language · Computer Science 2025-10-08 Harethah Abu Shairah , Hasan Abed Al Kader Hammoud , Bernard Ghanem , George Turkiyyah

Perturbation probing generates task-specific causal hypotheses for FFN neurons in large language models using two forward passes per prompt and no backpropagation, followed by a one-time intervention sweep of about 150 passes amortized…

Computation and Language · Computer Science 2026-05-01 Hongliang Liu , Tung-Ling Li , Yuhao Wu
‹ Prev 1 2 3 10 Next ›