Causal Interventions in Language Models

Follow this topic Ask about this topic

Momentum

55 papers in the last four weeks, up 162% on the four weeks before. 0.5% of all new papers.

Jul 13Week of Sep 28

Latest papers 317

All topics

Steering Interference Reflects the Model's Defaults, Not the Behavior Directions
Sep 7, 2026Srikanth Malla, Chiho Choi, Joon Hee ChoiLanguage Model Steering Causal Interventions in Language Models
Samsung Semiconductor US
Who Drives the Probability Game of VLMs? A Temporal Causal Drive Evaluation Framework
Sep 2, 2026Shuyao Xiao, Shengling Wang, Haoyu Niu +4VLM Evaluation VLM Interpretability
College of Artificial Intelligence, Beijing Normal University · Kuaishou Technology
Probing Factual Knowledge Transfer with Training Data Interventions
Sep 1, 2026Romina Oji, Marc Braun, Marcel Bollmann +2Language Model Pretraining Multilingual Language Model Evaluation
Department of Computer and Information Science Linköping University
Some Emotions Run Deeper: Layer-wise Probing and Causal Intervention in Large Language Models
Sep 1, 2026Tian Fang, Gaël Guibon, Davide BuscaldiCausal Interventions in Language Models Emotion Representation in Language Models
1Université Sorbonne Paris Nord, CNRS, Laboratoire d’Informatique de Paris Nord, LIPN · 2LORIA, CNRS, Université de Lorraine
CopyShield: A Cross-Level Benchmark of Copyright Defenses in LLMs
Sep 1, 2026Maryam Alshehyari, Dushyant Singh Chauhan, Samuele Poppi +3LLM Safety Benchmarks Direct Preference Optimization
Computing and Mathematical Science Division Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)
Hints Help But Do They Teach? Evaluating Skills Transfer in Code Generation
Sep 1, 2026Will BadrLanguage Model Generation Evaluation LLM Prompting
StateSwap: Probing Support-Elimination Hidden States in Multiple-Choice Questions
Sep 1, 2026Chao Gao, Haijiang Liu, Qiyuan Li +3Prompt Sensitivity Multiple-Choice Question Answering
Department of Computer Science, Vrije Universiteit Amsterdam, Amsterdam, The Netherlands · School of Computer Science and Technology, Wuhan University of Science and Technology, Wuhan, China · Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System, Wuhan University of Science and Technology, Wuhan, China
Lagged Coupling: Internal Representations Become Readable Before They Become Causal
Sep 1, 2026Xining XunCausal Interpretability Causal Interventions in Language Models
Tsingjiao Information Science (Beijing) Co., Ltd.
Context-Grounding Gains Are Mediated by Pre-existing Machinery: Auditing GRPO, SFT, and DPO
Sep 1, 2026Prakhar Gupta, Vaibhav GuptaRL for Language Models LLM Grounding
University of Michigan · University of Waterloo
How Do Language Models Choose Between Context and Memory?
Sep 1, 2026Benjamin Shih, John Winnicki, Arianna CaoKnowledge Conflicts in Language Models LLM Interpretability
Stanford University · Perpetual Labs
Latent Mechanisms of Language Control in Multilingual Language Models
Aug 31, 2026Ryo Mitsuhashi, Sabri Boughorbel, Majd HawaslyMultilingual Language Models Multilingual Language Model Evaluation
Princeton University New Jersey, USA · Prince Sattam bin Abdulaziz University Riyadh, Saudi Arabia · QCRI, Hamad Bin Khalifa University, Doha, Qatar
Asymmetries in Spontaneous and Instructed Deception
Aug 31, 2026Josiah LuikhamDeception in Language Models Causal Interventions in Language Models
Independent Researcher
The Safety Relay in Roleplay Jailbreaks: A Component-Resolved Causal Analysis of Harm Recognition and Refusal
Aug 31, 2026Md Mokarram Chowdhury, Ernie Chang, Yang LiJailbreak Attacks LLM Jailbreak Attacks
Department of Computer Science, Iowa State University, United States · Meta, United States
Enhancing Low-Resource Language Reasoning via High-Resource Language Feature Transfer
Aug 31, 2026Minju Song, Hyeon Hwang, Junhyun Lee +1Cross-Lingual Reasoning in Language Models Low-Resource Language Processing
Korea University · Hankuk University of Foreign Studies · Noah’s Farm +1
When Errors Become Memories: Causal Pathway Tracing in Multi-Turn Memory-Augmented LLMs
Aug 31, 2026Shuyao Xiao, Shengling Wang, Xuan Chen +8Memory-Augmented Language Models Causal Interventions in Language Models
College of Artificial Intelligence, Beijing Normal University, Beijing, China · 2Beike Language and Intelligence, Beijing, China · Center for Data Science, New York University, New York, NY, USA
Where You Measure Decides What You Measure: Position Selection in Ablation-Based SAE Evaluation
Aug 13, 2026Valentin NoëlSparse Autoencoders Causal Interventions in Language Models
Devoteam
Located but Not Releasable: Silent Gate Inversion and Bounded Linear Release
Aug 12, 2026Xining XunCausal Interventions in Language Models Activation Steering
Tsingjiao Information Science (Beijing) Co., Ltd.
Locating and Controlling Implicit Personalization in Large Language Models
Aug 12, 2026Yueru Yan, Siqi Wu, Thai LeLLM Personalization LLM Interpretability
Indiana University Bloomington, USA
Localizing Safety Alignment: MLP Layers and Mid-Network Blocks Encode Refusal Behavior in Large Language Models
Aug 12, 2026Mingyu Zong, Sampad Mohanty, Bhaskar KrishnamachariLLM Alignment LLM Interpretability
Department of Computer Science University of Southern California Los Angeles, CA 90007
Data Attribution of Emergent Misalignment with Persona Features
Aug 11, 2026Clemens Vetter, David Kaczér, Lucie Flek +1LLM Alignment Emergent Misalignment in Language Models
Bonn-Aachen International Center for Information Technology, University of Bonn, Germany
Every Token Counts: Exact Likert-Scale Distributions for Measuring LLM Attitudes and Biases
Aug 11, 2026Davood Wadi, Mohsen Ghodrat, Matthew PhilpLLM Evaluation Cultural Bias in Language Models
McGill University · University Canada West · Marketing Management +1
Decodable But Not Detachable: Training Data Granularity Determines Parametric Modularity in Large Language Models
Aug 10, 2026Marcus Armstrong, Navid Ayoobi, Arjun MukherjeeLLM Interpretability Circuits in Language Models
Department of Computer Science University of Houston Houston, TX 77204
What Would Fix This RAG Failure? Auditing Counterfactual Response with Paired Evidence Interventions
Aug 9, 2026Wenzhang DuRetrieval-Augmented Generation Counterfactual Reasoning in Language Models
Independent Researcher China
Thinking vs. NoThinking: Towards Interpreting Reasoning Mechanisms of Large Language Models via Sparse Autoencoders
Aug 8, 2026Bo Cheng, Qiaolin Lu, Yi Chang +1LLM Interpretability CoT Reasoning
School of Artificial Intelligence, Jilin University · The Hong Kong Polytechnic University · School of Artificial Intelligence, Jilin University Engineering Research Center of Knowledge-Driven Human-Machine Intelligence, MOE, China +1
When Is a Steerable Concept Representation Real? Measurement Confounds in a Cross-Family Audit of Neuroscience Parallels in LLMs
Aug 8, 2026Yuqi Wu, Shengming Zhao, Jie ChenLanguage Model Steering LLM Auditing
College of Biomedical Engineering, Fudan University
Decided Upstream, Written Late: Locating and Pricing the Cross-Lingual Refusal Circuit of a Multilingual MoE
Aug 8, 2026Ramakrishna P. Kompella, Aadit MahajanMultilingual Language Models Mixture-of-Experts Language Models
Gödel Machines · IIT Madras
From Behavior to Mechanism: Tracing Divergent Response Modes in Frontier Language Models
Aug 6, 2026Ali Jalal-KamaliLanguage Model Steering LLM Interpretability
USC Institute for Creative Technologies
The Illusion of Visual Tool-Use: A Causal Audit of Thinking with Images
Aug 6, 2026Zhiheng Wang, Bo Peng, Lai Wei +1Tool Use in VLMs Visually Grounded Reasoning
1Shanghai Jiao Tong University · 2Shanghai Artificial Intelligence Laboratory · 3Shanghai Innovation Institute
Signal or Spurious Cue? A Randomized Audit of Survey-Country Metadata in LLM Social Inference
Aug 6, 2026Yifan Lyu, Xinran Li, Jiaqi Qiao +1Probabilistic Forecasting Causal Interventions in Language Models
Dalian University of Technology, China
Recovering Lesion Parameters from Aphasic Picture Naming Error Profiles in Large Language Models
Aug 5, 2026Yong Yang, Roger Newman-Norlund, Xiang Guan +10LLM Interpretability Causal Interventions in Language Models
Department of Computer Science and Engineering, University of South Carolina · Department of Communication Sciences and Disorders, University of South Carolina · ALLT.AI, LLC +3

Sep 7, 2026, cs.LGJ/K move · Enter open · S save

Steering Interference Reflects the Model's Defaults, Not the Behavior Directions

Srikanth Malla, Chiho Choi, Joon Hee Choi

Samsung Semiconductor US

Language Model Steering Causal Interventions in Language Models Activation Steering

Open paper AskPDF arXiv

Activation steering promises modular control of language model behavior: a behavior such as politeness corresponds to a direction in a model's activations, and adding that direction while it generates should switch the behavior on and leave everything else alone. It does not. We ask what decides which other behaviors move, and by how much, and find that it is the model rather than the behavior being steered. A steer relaxes the model toward a small set of behaviors it already favors, chiefly refusal, sycophancy, and poeticism, and that set is much the same whatever is steered. Three results across 24 behaviors and ten instruction-tuned models support this, every effect read off the generated text by a language-model judge rather than off a probe. That readout matters: all 24 behaviors are linearly decodable, but only 20 change what the model writes. First, a direction carrying no behavioral content, matched to a real steer only in the size of the vector it adds, moves the same behaviors in the same order as real steers do, while producing none of the behaviors that need a specific direction. Second, most interference runs one way, so it cannot be an overlap between two directions: steering profanity makes the model toxic, while steering toxicity leaves profanity untouched. Third, with a behavior held out entirely, geometry measured on the others explains almost none of the interference it takes part in. The account holds on all ten models, the pull toward defaults strongest below 10B parameters and weakening in each family's largest. Reading a steer as a perturbation whose endpoint the model fixes implies that disentangling behavior directions cannot by itself make steering modular.

Momentum

Related topics

Latest papers 317