Inference-Time Language Model Alignment

Follow this topic Ask about this topic

The field studies how to guide a language model toward desired behavior at generation time, often without changing its weights. Approaches include prompt and persona conditioning, activation or logit steering, reward-guided decoding, and sampling or reranking methods such as Best-of-N. Researchers apply these methods to safety, helpfulness, reasoning, personalization, and conflicting preferences, while developing theoretical guarantees and more efficient ways to evaluate them. A central challenge is balancing gains in alignment against compute cost, reward-model errors, loss of useful capabilities, and the difficulty of transferring results across tasks and conversations.

Latest papers

78 papers13 papers in the last four weeks, up 8% on the four weeks before. 0.1% of all new papers.

ReMedI: Inference-Time Refinement and Self-Distillation for Controllable Molecular Optimization
Oct 23, 2026Jia Yi Zhang, Shivanshu Shekhar, Allison McLevige +1Molecular Optimization Inference-Time Language Model Alignment
University of Illinois Urbana-Champaign
Model vs. User: Whose Values Prevail? Inference-Time Value Alignment in Multi-Turn Conversations
Oct 14, 2026Enrico Liscio, Anne Arzberger, Aditya Shankar +1LLM Alignment Inference-Time Language Model Alignment
Delft University of Technology, the Netherlands
Bayesian Test-Time Inference of Task-Aligned Similarity from Weak Interactive Feedback
Oct 13, 2026Weida Liang, Kenji KawaguchiInference-Time Language Model Alignment Bayesian Inference
National University of Singapore
Efficient Best-of-N policy evaluation for inference-time alignment
Oct 7, 2026Jonas Schweisthal, Yuxin Wang, Athiya Deviyani +2Best-of-N Sampling Off-Policy Evaluation
LMU Munich & MCML · Carnegie Mellon University · Meta (work in personal capacity)
Metropolis-Hastings Dominates Importance Resampling for Policy Composition
Oct 2, 2026Alexey Kurennoy, Ramil Yarullin, Fergal ReidIndependent Metropolis-Hastings Language Model Decoding
DNAlign: Dynamic Null-Space Safe Alignment for LLMs
Oct 2, 2026Jisheng Dang, Yushuo Zhao, Dewei Liu +6LLM Alignment LLM Safety Alignment
School of Information Science and Engineering, Lanzhou University, Lanzhou 730000, China · National University of Singapore, Singapore 119077 · School of Medical Technology, Beijing Institute of Technology, Beijing 100081, China
The Asymptotics of Language Model Alignment with Memory
Oct 1, 2026Haricharan Balasundaram, V. Arvind RameshwarRL for Language Models LLM Alignment
School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA · IIT Madras · Department of Electrical Engineering, IIT Madras, Chennai, India
HeadEdit: Calibrating Language Model Behavior Through the Frozen Unembedding Matrix
Oct 1, 2026Zirui He, Haiyan Zhao, Jingyu Hu +5Language Model Steering Language Model Calibration
New Jersey Institute of Technology · University of Bristol · Kuaishou Technology +1
Ready2Blend: From Natural-Language Instructions to Composable Alignment Prompts
Sep 30, 2026Jeesu Jung, Hwan Chang, Juseon Do +5Continual Learning for LLMs LLM Alignment
Korea Advanced Institute of Science and Technology · Samsung SDS
Demographic Pluralism: Inference-Time Modeling of Pluralistic Human Preference Distributions
Sep 29, 2026Meng-Chen Wu, Qipin Chen, Ansh Jain +3Human Preference Modeling Inference-Time Language Model Alignment
Amazon Science
Aligned Data Can Induce Misalignment via Context Confusion
Sep 29, 2026Yavuz Bakman, Duygu Nur Yaldiz, Baris Askin +4LLM Alignment LLM Post-Training
University of Southern California · Carnegie Mellon University · Amazon +1
Constitutional adapters: Inference-time interventions for misalignment and misuse
Sep 29, 2026Adam S. Lowet, Mark KurzejaLLM Safety Alignment LLM Jailbreak Attacks
Anthropic Fellows Program · Independent
Rethinking Personalized Generation: Test-Time Alignment via Factorized Ranking Models
Sep 28, 2026Qiyao Ma, Junshan Zhang, Zhe ZhaoLearning to Rank LLM Reranking
University of California, Davis
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
Minimally Invasive Steering of Language Models
Sep 24, 2026Taha Entesari, Jingyu Zhang, Daniel Khashabi +1Language Model Steering Inference-Time Language Model Alignment
Johns Hopkins University
From Static Personal Values to Contextualized Personalization: Bayesian Personalized Value Alignment for LLMs
Sep 24, 2026Hanze Guo, Aixuan Song, Jing Yao +4LLM Alignment Personalized Language Models
Gaoling School of Artificial Intelligence, Renmin University of China · Microsoft Research Asia · Beijing Key Laboratory of Research on Large Models and Intelligent Governance +1
ExpBoN: Exponential-Noise Best-of- $n$ for Efficient Test-Time LLM Alignment
Sep 18, 2026Yanxiao Liu, Sicheng Wan, Deniz GündüzBest-of-N Sampling Inference-Time Language Model Alignment
Imperial College London · University of Washington
An Efficient and Modular Framework for Targeted Harm Mitigation in LLMS
Sep 12, 2026Roberto Campbell, Momin Abbas, Muneeza Azmat +5LLM Safety Alignment Low-Rank Adaptation
University of California, Santa Cruz · IBM Research
Inference-Time Nash Alignment
Sep 8, 2026Hadi Hosseini, Debmalya Mandal, Duohan ZhangInference-Time Optimization Preference Alignment
Penn State University, USA · University of Warwick, UK
Cross-Lingual Representation Alignment by Token-Level Optimal Transport in a Language-Agnostic Space
Sep 6, 2026Taisei Yamamoto, Ryoma Kumon, Danushka Bollegala +1Cross-Lingual Representation Alignment Cross-Lingual Consistency in Language Models
The University of Tokyo · Riken · University of Liverpool +1
Spillover-Aware Multi-Value Steering for Pluralistic LLM Alignment
Sep 5, 2026Weici Pan, Xander Barron, Jiawei Zhou +1Language Model Steering Inference-Time Language Model Alignment
Stony Brook University
Instruction Duplication as an Inference-Time Control Primitive
Sep 3, 2026Victor LavrenkoInstruction Following Inference-Time Language Model Alignment
PeaceTech VC, Israel
Swiss-Knife: A Framework for Reconfigurable Externalised Multi-Objective Alignment at Decode Time
Sep 3, 2026Agnibh Karmakar, Mayur Parvatikar, Shreyash Dhoot +5Multi-Objective Language Model Alignment Inference-Time Language Model Alignment
BITS Pilani Hyderabad Campus, India · RV College of Engineering, India · Pragya Lab, BITS Pilani Goa Campus, India +4
Trust Your Guide Only When Certain: Uncertainty-Aware Sparse Alignment at Inference Time
Sep 1, 2026Zeen Zhu, Zhuo Li, Weiyang Guo +4LLM Safety Alignment Inference-Time Language Model Alignment
Harbin Institute of Technology, Shenzhen, China · Huawei Technologies Co., Ltd. · Beijing Academy of Artificial Intelligence, China.
Beyond Token-Level Guidance: Inference-Time Alignment of Specialized LLMs via Cross-Family Representation Steering
Aug 31, 2026Jin Gan, Xin Li, Jun LuoLLM Safety Alignment Language Modeling
College of Computing and Data Science, Nanyang Technological University, Singapore · College of Cryptology and Cyber Science, Nankai University, Tianjin, China
Evolutionary Soups: Evolving Mixture-of-Experts for Multi-Objective LLM Alignment
Aug 30, 2026Lingxiao Kong, Steffen Staab, Cong Yang +2Mixture-of-Experts Language Models Multi-Objective Language Model Alignment
Fraunhofer Institute for Applied Information Technology FIT · University of Cologne · University of Stuttgart +3
Large Language Models Systematically Favor Popular Options: Evidence and Mitigation Across MCQs
Aug 29, 2026Abdelrahman Abdallah, Mohammed Ali, Bhawna Piryani +2Popularity Bias Language Model Bias Evaluation
University of Innsbruck · Chungbuk National University
Privacy Without Regret: Differentially Private Inference-Time Alignment
Aug 26, 2026Ishi Jain, Nandini Bhattad, Sayak Ray ChowdhuryInference-Time Language Model Alignment Best-of-N Sampling
Indian Institute of Technology Kanpur
ST $^2$ U: Stateful Test-Time Unlearning via Restricted Knowledge Boundary Control
Aug 24, 2026Xunlei Chen, Qinghui Gong, Ruini Xue +3LLM Unlearning Machine Unlearning
University of Electronic Science and Technology of China · Southwest Jiaotong University
Latent Ordinal Evidence, Misaligned Outputs: Inference-Time Ordinal Lens Alignment for Multimodal LLMs
Aug 21, 2026Haiming Li, Yingsheng Liu, Jingmin Zhu +5Multimodal Large Language Models Ordinal Regression
Faculty of Information Technology, Monash University, Melbourne, Australia · Monash University, Melbourne, Victoria, Australia

Oct 23, 2026J/K move · Enter open · S save

ReMedI: Inference-Time Refinement and Self-Distillation for Controllable Molecular Optimization

Jia Yi Zhang, Shivanshu Shekhar, Allison McLevige, Tong Zhang

University of Illinois Urbana-Champaign

Molecular Optimization Inference-Time Language Model Alignment Self-Training for Language Models

Open paper AskPDF arXiv

In drug development, lead optimization must improve competing properties, such as increasing permeability and reducing toxicity, while limiting scaffold changes that could compromise the lead's potency. RL-based post-training, such as C-MORAL, addresses this problem by fine-tuning molecular language models to directly optimize molecular leads for objectives of interest, which relies on expensive task-specific datasets and faces challenges in achieving multi-objective optimization (MOO). We argue a pre-trained model may already contain useful molecular edits that zero-shot generation cannot efficiently reach. Motivated by the limited search capabilities of zero-shot generation, we develop an efficient Metropolis-Hastings search strategy to better explore the proposal distribution. We propose ReMedI, which uses the molecular language model as a base proposer for inference-time alignment. Given a lead and a set of multi-objective tasks, the model proposes edits conditioned on the current molecule, while Metropolis--Hastings refinement guides sequential edits under an objective for property improvement, preservation, and lead similarity. On both C-MuMOInstruct and SMDD benchmarks, refinement improves frozen base models across three backbones, reaching significantly higher multi-objective optimization success rates compared to baselines. These search traces can then distilled into the base model which improves zero-shot generation and subsequent search using approximately 7.8 times less training molecular leads than that of C-MORAL. ReMedI achieves approximately 2.8 times the strict-success coverage of RL baseline C-MORAL, including on out-of-distribution tasks, with matched lead-relative similarity.