English
Related papers

Related papers: Feature-Guided SAE Steering for Refusal-Rate Contr…

200 papers

While the activations of neurons in deep neural networks usually do not have a simple human-understandable interpretation, sparse autoencoders (SAEs) can be used to transform these activations into a higher-dimensional latent space which…

Machine Learning · Computer Science 2025-08-07 Gonçalo Paulo , Alex Mallen , Caden Juang , Nora Belrose

Large language models (LLMs) have recently been employed as agents to solve sequential decision-making tasks such as Bayesian optimization and multi-armed bandits (MAB). These works usually adopt an LLM for sequential action selection by…

Machine Learning · Computer Science 2025-02-04 Mingze Kong , Zhiyong Wang , Yao Shu , Zhongxiang Dai

Large Language Models (LLMs) have emerged as powerful tools, but their inherent safety risks - ranging from harmful content generation to broader societal harms - pose significant challenges. These risks can be amplified by the recent…

Prepending model inputs with safety prompts is a common practice for safeguarding large language models (LLMs) against queries with harmful intents. However, the underlying working mechanisms of safety prompts have not been unraveled yet,…

Machine Learning · Computer Science 2024-06-04 Chujie Zheng , Fan Yin , Hao Zhou , Fandong Meng , Jie Zhou , Kai-Wei Chang , Minlie Huang , Nanyun Peng

Sparse autoencoders (SAEs) improve interpretability in multimodal models, but it remains unclear whether SAE features form modular, composable units for reasoning-an assumption underlying many intervention-based steering methods. We test…

Artificial Intelligence · Computer Science 2026-03-27 Yunpeng Zhou

Improving Sparse Autoencoders (SAEs) requires benchmarks that can precisely validate architectural innovations. However, current SAE benchmarks on LLMs are often too noisy to differentiate architectural improvements, and current synthetic…

Machine Learning · Computer Science 2026-02-17 David Chanin , Adrià Garriga-Alonso

Large Language Models (LLMs) have showcased remarkable capabilities in following human instructions. However, recent studies have raised concerns about the robustness of LLMs when prompted with instructions combining textual adversarial…

Computation and Language · Computer Science 2024-02-27 Yuansen Zhang , Xiao Wang , Zhiheng Xi , Han Xia , Tao Gui , Qi Zhang , Xuanjing Huang

In autonomous driving, end-to-end (E2E) driving systems that predict control commands directly from sensor data have achieved significant advancements. For safe driving in unexpected scenarios, these systems may additionally rely on human…

Computer Vision and Pattern Recognition · Computer Science 2026-03-17 Seo Hyun Kim , Jin Bok Park , Do Yeon Koo , Hogun Park , Il Yong Chun

Understanding the internal representations of large language models is crucial for ensuring their reliability and safety, with sparse autoencoders (SAEs) emerging as a promising interpretability approach. However, current SAE training…

Machine Learning · Computer Science 2025-10-13 T. Ed Li , Junyu Ren

Predicting protein function from amino acid sequence remains a central challenge in data-scarce (low-$N$) regimes, limiting machine learning-guided protein design when only small amounts of assay-labeled sequence-function data are…

Machine Learning · Computer Science 2025-08-27 Darin Tsui , Kunal Talreja , Amirali Aghazadeh

What latent features are encoded in language model (LM) representations? Recent work on training sparse autoencoders (SAEs) to disentangle interpretable features in LM representations has shown significant promise. However, evaluating the…

The prevailing approach to aligning Large Language Models (LLMs) typically relies on human or AI feedback and assumes access to specific types of preference datasets. In our work, we question the efficacy of such datasets and explore…

Machine Learning · Computer Science 2024-03-19 Hao Sun

Activation steering is a promising technique for controlling LLM behavior by adding semantically meaningful vectors directly into a model's hidden states during inference. It is often framed as a precise, interpretable, and potentially…

Machine Learning · Computer Science 2026-02-17 Anton Korznikov , Andrey Galichin , Alexey Dontsov , Oleg Y. Rogov , Ivan Oseledets , Elena Tutubalina

Mixture-of-Experts (MoE) in Large Language Models (LLMs) routes each token through a subset of specialized Feed-Forward Networks (FFN), known as experts. We present SteerMoE, a framework to steer MoE models by detecting and controlling…

Computation and Language · Computer Science 2026-02-13 Mohsen Fayyaz , Ali Modarressi , Hanieh Deilamsalehy , Franck Dernoncourt , Ryan Rossi , Trung Bui , Hinrich Schütze , Nanyun Peng

Aligning large language models (LLMs) with human expectations without human-annotated preference data is an important problem. In this paper, we propose a method to evaluate the response preference by using the output probabilities of…

Computation and Language · Computer Science 2024-08-16 Aiwei Liu , Haoping Bai , Zhiyun Lu , Xiang Kong , Simon Wang , Jiulong Shan , Meng Cao , Lijie Wen

Fine-tuning Large Language Models (LLMs) has emerged as a common practice for tailoring models to individual needs and preferences. The choice of datasets for fine-tuning can be diverse, introducing safety concerns regarding the potential…

Computation and Language · Computer Science 2024-10-15 Hyeong Kyu Choi , Xuefeng Du , Yixuan Li

Sparse autoencoders (SAEs) are a popular tool for interpreting large language model activations, but their utility in addressing open questions in interpretability remains unclear. In this work, we demonstrate their effectiveness by using…

Machine Learning · Computer Science 2025-04-21 Dmitrii Kharlapenko , Stepan Shabalin , Fazl Barez , Arthur Conmy , Neel Nanda

While modern LLMs are aligned to refuse harmful requests, it is essential to understand the underlying mechanistic basis of this refusal behavior for model safety analysis. For example, steering-based jailbreak attacks exploit this by…

Artificial Intelligence · Computer Science 2026-05-28 Su-Hyeon Kim , Hyundong Jin , Yejin Lee , Yo-Sub Han

Understanding what features are encoded by learned directions in LLM activation space requires identifying inputs that strongly activate them. Feature visualization, which optimizes inputs to maximally activate a target direction, offers an…

Machine Learning · Computer Science 2026-02-23 João N. Cardoso , Arlindo L. Oliveira , Bruno Martins

System prompts provide a lightweight yet powerful mechanism for conditioning large language models (LLMs) at inference time. While prior work has focused on English-only settings, real-world deployments benefit from having a single prompt…

Computation and Language · Computer Science 2025-12-03 Lechen Zhang , Yusheng Zhou , Tolga Ergen , Lajanugen Logeswaran , Moontae Lee , David Jurgens