Causal Interventions in Language Models

Follow this topic Ask about this topic

This field asks which internal states, features, heads, neurons, or context elements actually influence a language model’s decisions, rather than merely correlate with them. Researchers use counterfactual prompt and evidence edits, activation patching and ablation, causal tracing, sparse-autoencoder features, and inference-time steering or weight edits to test mechanisms and control behavior. Applications include checking whether chain-of-thought and retrieved evidence drive answers, and modifying refusal, hallucination, bias, tool use, or reasoning while preserving unrelated capabilities. Results show that readable representations are often not causally used or easily steered; effects vary by layer, task, model, and context, while redundancy and collateral effects make reliable, selective interventions a continuing challenge.

Latest papers

489 papers83 papers in the last four weeks, up 22% on the four weeks before. 0.5% of all new papers.

Debiasing Reward Models via Causally Motivated Inference-Time Intervention
Apr 30, 2026Kazutoshi Shinoda, Kosuke Nishida, Kyosuke NishidaReward Modeling LLM Alignment
Human Informatics Labs., NTT, Inc.
Perturbation Probing: A Two-Pass-per-Prompt Diagnostic for FFN Behavioral Circuits in Aligned LLMs
Apr 30, 2026Hongliang Liu, Tung-Ling Li, Yuhao WuLLM Interpretability Mechanistic Interpretability
Palo Alto Networks
Compliance versus Sensibility: On the Reasoning Controllability in Large Language Models
Apr 29, 2026Xingwei Tan, Marco Valentino, Mahmud Elahi Akhter +3LLM Interpretability Instruction Following
School of Computer Science, University of Sheffield · School of EECS, Queen Mary University of London · The Alan Turing Institute
How a Cooperative-Override Circuit Suppresses Nash Play in Large Language Models
Apr 29, 2026Paraskevas V. Lekeas, Giorgos StamatopoulosNash Equilibrium Game Theory
DreamWorks Animation, Glendale, USA · Department of Economics, University of Crete, Rethymno, Crete, Greece
MoRFI: Monotonic Sparse Autoencoder Feature Identification
Apr 29, 2026Dimitris Dimakopoulos, Shay B. Cohen, Ioannis KonstasLLM Hallucination Mitigation LLM Interpretability
University of Edinburgh, UK · Heriot-Watt University, UK
reward-lens: A Mechanistic Interpretability Library for Reward Models
Apr 28, 2026Mohammed Suhail B NadafReward Modeling LLM Interpretability
Why Are Some Emotions Harder for LLMs? Uncovering the Causal Mechanisms of Emotion Inference via Sparse Autoencoders
Apr 28, 2026Bangzhao Shu, Arinjay Singh, Mai ElSheriefLLM Interpretability Causal Interventions in Language Models
Northeastern University
Authority Inversion in LLM-Mediated Ubiquitous Systems: When Models Trust Users Over Sensors
Apr 28, 2026Long Zhang, Zi-bo Qin, Wei-neng ChenKnowledge Conflicts in Language Models LLM Auditing
School of Computer Science and Engineering, South China University of Technology, Guangzhou, China
Pruning via Causal Attribution Preserves Reasoning Performance in Large Language Models
Apr 27, 2026Amogh Sheth, Andrew Lin, Biruk Assefa +2LLM Pruning Self-Attention
Edison Academy Magnet School · Massachusetts Institute of Technology · State University of New York College at Plattsburgh +2
When Chain-of-Thought Fails, the Solution Hides in the Hidden States
Apr 25, 2026Houman Mehrafarin, Amit Parekh, Ioannis KonstasTransformer Interpretability CoT Reasoning
Heriot-Watt University, Edinburgh, United Kingdom · University of Edinburgh, Edinburgh, United Kingdom
Preference Heads in Large Language Models: A Mechanistic Framework for Interpretable Personalization
Apr 24, 2026Weixu Zhang, Ye Yuan, Changjiang Han +7LLM Alignment Language Model Steering
McGill University · Mila - Quebec AI Institute · MBZUAI +2
Self-Correction as Feedback Control: Error Dynamics, Stability Thresholds, and Prompt Interventions in LLMs
Apr 24, 2026Aofan Liu, Jingxiang MengLLM Self-Correction Language Model-Based Control
1Peking University. · University of Chicago.
Dissociating Decodability and Causal Use in Bracket-Sequence Transformers
Apr 24, 2026Aryan Sharma, Cutter Dawes, Shivam RavalTransformer Interpretability Self-Attention
1Yale University · 2Independent · 3Harvard University
Dual-Stance Evaluation of Sycophancy: The Structure of Agreement and the Limits of Intervention
Apr 22, 2026Matthew James BuchanLLM Sycophancy Language Model Steering
Independent Researcher, United Kingdom.
Where Reasoning Breaks: Logic-Aware Path Selection by Controlling Logical Connectives in LLMs Reasoning Chains
Apr 22, 2026Seunghyun Park, Yuanyuan LeiInference-Time Search RL for Language Model Reasoning
Independent Researcher · University of Florida Gainesville, FL, United States
Separable Pathways for Causal Reasoning: How Architectural Scaffolding Enables Hypothesis-Space Restructuring in LLM Agents
Apr 21, 2026John Alderete, Sebastian Benthal, Connie Xu +1Causal Discovery LLM Agents
Amigo AI · Simon Fraser University · International Computer Science Institute
Are LLM Uncertainty and Correctness Encoded by the Same Features? A Functional Dissociation via Sparse Autoencoders
Apr 21, 2026Het Patel, Tiejin Chen, Hua Wei +2LLM Interpretability LLM Reliability
University of California, Riverside, CA · Arizona State University, Tempe, AZ
LLMs Know They're Wrong and Agree Anyway: The Shared Sycophancy-Lying Circuit
Apr 21, 2026Manav PandeyLLM Sycophancy LLM Alignment
Georgia Institute of Technology
Latent Phase-Shift Rollback: Inference-Time Error Correction via Residual Stream Monitoring and KV-Cache Steering
Apr 20, 2026Manan Gupta, Dhruv KumarLLM Self-Correction Language Model Error Detection
1BITS Pilani, Pilani Campus, India
State Transfer Reveals Reuse in Controlled Routing
Apr 20, 2026Yanzhen Lu, Zhicheng Qian, Muchen Jiang +1Prompt Learning Transfer Learning
Shanghai Qizhi Institute · Southeast University · Nanjing Whale Cloud +1
Are Emotion and Rhetoric Neurons in LLM? Neuron Recognition and Adaptive Masking for Emotion-Rhetoric Prediction Steering
Apr 19, 2026Li Zheng, Xin Zhang, Shuyi He +5Language Model Steering LLM Interpretability
Key Laboratory of Aerospace Information Security and Trusted Computing, Ministry of Education, School of Cyber Science and Engineering, Wuhan University, Wuhan, China · Ant International · School of Computing Technologies, RMIT University, Australia
One Mask to Rule Them All: On Hidden Facts after Editing and How to Find Them
Apr 18, 2026Ali Holmov, Paul Youssef, Nandi Schoots +1Knowledge Editing Transformer Interpretability
Technical University of Munich · Marburg University · University of Oxford
Predicting Where Steering Vectors Succeed
Apr 16, 2026Jayadev BillaLanguage Model Steering Representation Steering
CausalDetox: Causal Head Selection and Intervention for Language Model Detoxification
Apr 16, 2026Yian Wang, Yuen Chen, Agam Goyal +1Causal Reasoning in Language Models LLM Safety
Department of Computer Science University of Illinois, Urbana-Champaign Champaign, IL 61801
Mechanistic Decoding of Cognitive Constructs in Large Language Models
Apr 16, 2026Yitong Shou, Manhao GuanLLM Interpretability Causal Interventions in Language Models
Zhejiang University, Hangzhou 310027, China. · The authors are with the College of Computer Science and Technology, Zhejiang University, Hangzhou 310027, China.
Weight Patching: Toward Source-Level Mechanistic Localization in LLMs
Apr 15, 2026Chenghao Sun, Chengsheng Zhang, Guanzheng Qin +2Mechanistic Interpretability Causal Interventions in Language Models
Key Laboratory of Ministry of Education for Brain Inspired Intelligent Perception and Cognition, University of Science and Technology of China, Hefei, Anhui 230026, China
Psychological Concept Neurons: Can Neural Control Bias Probing and Shift Generation in LLMs?
Apr 13, 2026Yuto Harada, Hiro Taiyo HamadaPersonality Modeling in Language Models LLM Interpretability
Araya Inc.
CausalGaze: Unveiling Hallucinations via Counterfactual Graph Intervention in Large Language Models
Apr 13, 2026Linggang Kong, Lei Wu, Yunlong Zhang +4LLM Hallucination Detection Hallucination in Language Models
College of Electronic Engineering, National University of Defense Technology · Anhui Province Key Laboratory of Cyberspace Security Situation Awareness and Evaluation · Institute of Computer Application, China Academy of Engineering Physics
Computational Lesions in Multilingual Language Models Separate Shared and Language-specific Brain Alignment
Apr 12, 2026Yang Cui, Jingyuan Sun, Yizheng Sun +8Neural Encoding Multilingual Language Models
Department of Computer Science, The University of Manchester, Manchester, UK. · State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, CAS, Beijing, China. · School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China. +3
CoSToM:Causal-oriented Steering for Intrinsic Theory-of-Mind Alignment in Large Language Models
Apr 11, 2026Mengfan Li, Xuanhua Shi, Yang DengTheory of Mind LLM Alignment
National Engineering Research Center for Big Data Technology and System, Services Computing Technology and System Lab, Cluster and Grid Computing Lab, Huazhong University of Science and Technology · Singapore Management University

Apr 30, 2026J/K move · Enter open · S save

Debiasing Reward Models via Causally Motivated Inference-Time Intervention

Kazutoshi Shinoda, Kosuke Nishida, Kyosuke Nishida

Human Informatics Labs., NTT, Inc.

Reward Modeling LLM Alignment Debiased ML Language Model Bias Evaluation Causal Interventions in Language Models

Open paper AskPDF arXiv

Reward models (RMs) play a central role in aligning large language models (LLMs) with human preferences. However, RMs are often sensitive to spurious features such as response length. Existing inference-time approaches for mitigating these biases typically focus exclusively on response length, resulting in performance trade-offs. In this paper, we propose causally motivated intervention for mitigating multiple types of biases in RMs at inference time. Our method first identifies neurons whose activations are strongly correlated with predefined bias attributes, and applies neuron-level intervention that suppresses these signals. We evaluate our method on RM benchmarks and observe reductions in sensitivity to spurious features across diverse bias types, without inducing performance trade-offs. Moreover, when used for preference annotation, small RMs (2B and 7B) with our method, which edits less than 2% of all the neurons in RMs, enable LLMs to improve alignment, achieving performance comparable to that of a state-of-the-art 70B RM on AlpacaEval and MT-Bench. Further analysis reveals that bias signals are primarily encoded by neurons in early layers, shedding light on the internal mechanisms of bias exploitation in RMs.