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

Scalable Attribution and Control of Model Behavior During Training
Sep 27, 2026Sleem AbdelghafarGradient-Based Attribution Causal Interventions in Language Models
Rice University
When Does the Concept of "Dog" Emerge in an Audio LLM?
Sep 27, 2026Zhe Wang, Shiqi Liu, Ruiyun Zhong +2Audio Understanding Audio-Language Models
Huazhong University of Science and Technology, Wuhan, China
Decoupling Token Roles in Autoregressive Pretraining
Sep 27, 2026Suqin Yuan, Runqi Lin, Kevin Qinghong Lin +4Language Model Pretraining Autoregressive Language Modeling
University of Sydney · University of Oxford · Southeast University
Direct Hidden-State Alignment: Mapping and Controlling Preference Expression in LLMs
Sep 27, 2026Fansheng Zhang, Shengran Guo, Zexiao Wang +3LLM Alignment Preference Alignment
Chengdu University · North Carolina State University · Fudan University +2
Does a model's stated reason for rejecting a candidate do any work?
Sep 24, 2026Archit RastogiFaithfulness of Language Model Explanations Causal Interventions in Language Models
Independent Researcher
Thinking Leakage: A Causal Audit of NoThink Post-Training in Hybrid Reasoning Models
Sep 23, 2026Zehao Liu, Vasant G. HonavarLLM Post-Training Causal Interventions in Language Models
College of Information Sciences and Technology Pennsylvania State University
Scaling Attention Head Analysis via Gradient-Based Attribution in Context-Aware Machine Translation
Sep 23, 2026Paweł Mąka, Yusuf Can Semerci, Jan Scholtes +1Gradient-Based Attribution Attention Head Analysis
Department of Advanced Computing Sciences Maastricht University
From Decorative to Load-Bearing: Task Difficulty Shapes the Causal Role of Chain-of-Thought
Sep 21, 2026Renee Jia, Di MuCoT Faithfulness Causal Interventions in Language Models
R2M AI, Cornell University · R2M AI, Carnegie Mellon University
Attention as a Routing Graph: Live Circuit Extraction from a Single Forward Pass
Sep 21, 2026Ash Manvi, Samreena TajreenCircuit Discovery Attention Mechanisms
Aquin Labs
Circuit-Diff: Factual Edit-based Intervention Method for Localizing Knowledge in Attribution Graphs
Sep 20, 2026Edward G. Friedman, Xiangchen SongCircuit Discovery Mechanistic Interpretability
School of Computer Science Machine Learning Department Carnegie Mellon University Pittsburgh, PA 15213
From Concept Alignment to Causal Grounding: An Intervention Test of Chain-of-Thought Faithfulness
Sep 19, 2026Qianli Wang, Yilong Wang, Dennis Wei +5CoT Faithfulness LLM Interpretability
Technische Universität Berlin · German Research Center for Artificial Intelligence (DFKI) · IBM Research +4
Deep Noir: Autonomous Steering Discovery via Architectural Chronometry in Transformer Models
Sep 17, 2026Frank E. Bobe, Gregory D. Vetaw, Darshan W. Bryner +2Transformer Interpretability Language Model Steering
Naval Surface Warfare Center Panama City Division
Planning or Improvisation? Stress-Testing the Poetry Planning Site on Open Models and Open Cross-Layer Transcoders
Sep 16, 2026Éric JacopinLLM Planning Mechanistic Interpretability
Cosmic AI, France
Decodability is Not Causality: Dissociating Probe Readouts from Behavioral Drivers via SAE Decomposition
Sep 16, 2026Devesh Tiwari, Camille Davis, Shivank Sinha +3Causal Interventions in Language Models Hidden-State Probing
WW-P High School South · Phillips Academy Andover · Dublin High School +3
TAME: Token Attribution and Masking for Emergent misalignment
Sep 15, 2026Md Rayhanul Masud, Md Rizwan ParvezLLM Interpretability LLM Fine-Tuning
University of California, Riverside · 2Qatar Computing Research Institute (QCRI)
Empathy Is Steerable but Multi-Axial: Mechanism Geometry and Persona Effects in LLMs
Sep 14, 2026JuHeon Ha, Byounghan Lee, Yunseo Choi +1Language Model Steering LLM Interpretability
Department of Artificial Intelligence, Ajou University
Refusal Reads Only a Slice of What the Model Knows: Harm-Keyed Routing and Its Exceptions Across Model Families
Sep 13, 2026Orion Reblitz-RichardsonLLM Alignment Moral Reasoning in Language Models
Distiller Labs.
Calibrating Interpretability Instruments Before Trusting Their Verdicts
Sep 13, 2026Orion Reblitz-RichardsonLLM Interpretability Causal Interpretability
Distiller Labs
From Parameters to Answers: How LLMs Retrieve and Use Their Internal Knowledge
Sep 12, 2026Wenkang Wei, Yuan Fang, Renhe Jiang +2LLM Interpretability Factual Knowledge in Language Models
University of Science and Technology of China Hefei, China · Singapore Management University Singapore, Singapore · The University of Tokyo Tokyo, Japan +1
Generative Interpretability via Scalable Neuro-Symbolic Models
Sep 11, 2026Xiaocong YangLLM Interpretability LLM Safety
Causal Analysis and Mitigation of Spurious Onsets in Full-Duplex Speech LLMs
Sep 11, 2026Kento NishiSpeech Language Models Causal Interventions in Language Models
Massachusetts Institute of Technology · Comcast Applied AI Research, Speech AI Team
K/V-Cache Interventions Dissociate Representation Alignment from Persona Expression in Decoder-Only Language Models
Sep 11, 2026Yu Sun, Mengyin Lu, Cong Feng +2KV Caching Decoder-Only Language Models
LinkedIn
The information geometry of large language models is shared, learned, and controllable
Sep 11, 2026Dario PicozziInformation Geometry Language Model Steering
Department of Physics and Astronomy, University College London (UCL), Gower Street, London WC1E 6BT, United Kingdom · London Centre for Nanotechnology, 19 Gordon St, London WC1H 0AH, United Kingdom
MUtE: A Dual Framework for Concept Erasure and Counterfactual Interventions
Sep 11, 2026Antoine SaillenfestCausal Counterfactual Generation Causal Interventions in Language Models
onepoint, 29 rue des Sablons, 75116 Paris (France)
Through the Looking Glass: Directly Reading and Writing Transformers
Sep 9, 2026Mark OskinKnowledge Editing Transformer Interpretability
School of Computer Science and Engineering University of Washington
Record Grouping Controls Evidence Weight in Language Models
Sep 8, 2026Zhongxuan Liu, Sicheng Zhou, Hongzhi WangCausal Interventions in Language Models
Faculty of Computing, Harbin Institute of Technology
Target-Independent Micro-Interventions for Predicting Training Response Across Language-Model Families
Sep 8, 2026Zhongxuan Liu, Sicheng Zhou, Hongzhi WangCausal Interventions in Language Models
Faculty of Computing, Harbin Institute of Technology
Key Path Identification for Resolving Knowledge Conflicts via SAE-based Steering
Sep 8, 2026Wenbo Zhang, Zhongxiang Sun, Zhiguang Han +1Language Model Steering Knowledge Conflicts in Language Models
Gaoling School of Artificial Intelligence, Renmin University of China · Nanyang Technological University
InfluenceField: A Differentiable Field with Interventionally Identifiable Causal Structure for Multimodal World Modeling
Sep 7, 2026Zihao Yang, Zijia Wang, Zhiqiu HuangCausal Representation Learning Causal World Models
Mathematical Institute, University of Oxford · Department of Computer Science, University of Oxford · School of Mathematical Sciences, University of Nottingham Ningbo China
LLM Forensics: Where Do Backdoors Hide? Localizing and Controlling Trigger Mechanisms with Sparse Autoencoders
Sep 7, 2026Wissam Antoun, Francis Kulumba, Théo Lasnier +2LLM Backdoor Attacks Circuits in Language Models
Inria Paris

Sep 27, 2026, cs.LGJ/K move · Enter open · S save

Scalable Attribution and Control of Model Behavior During Training

Sleem Abdelghafar

Rice University

Gradient-Based Attribution Causal Interventions in Language Models Neural Network Training Dynamics Training Data Attribution

Open paper AskPDF arXiv

Attributing and controlling model behavior during training requires identifying each example's contribution quickly enough to act before the next update. However, examples in the same training batch can produce similar behavioral changes, making their individual contributions difficult to distinguish. We address this ambiguity through mutual information, accounting for interference within the batch by quantifying how much the combined behavioral change reveals about each example's contribution. We show that this mutual information is a logarithmic function of Behavioral Gradient Uniqueness (BGU). BGU gives the information measure its geometric interpretation. Our Batch-Space Ghost (BS-Ghost) algorithm makes these scores practical inside the training loop through shared computation in batch space, without storing model-sized example gradients. On a complete 1,000-example Qwen2.5-7B-Instruct workload, our BS-Ghost implementation adds 27 seconds (8.0%) to 5.5 minutes of ordinary training. Removal and retraining demonstrate that BGU identifies data that causally shapes final behavior. At each training step, signed information identifies which examples strengthen or weaken the target behavior, explaining how behavior develops during training. Signed information also enables cheap intervention during training: it predicts how changing example weights will affect behavior in the next update. We then use these predictions to choose weights that steer behavior toward a desired target. This makes our framework a practical foundation for scalable oversight and verification of training pipelines and processes, helping evaluators assess model alignment, understand how it develops during training, and guide interventions that shape ongoing learning.

Momentum

Related topics

Latest papers 317