LLM Alignment

LLM: Large Language Model

Follow this topic Ask about this topic

LLM alignment studies how to shape language models’ behavior so they are helpful, truthful, safe, and responsive to the people and settings they serve. Common approaches include supervised fine-tuning, preference learning such as Direct Preference Optimization, reinforcement learning from human or AI feedback, and inference-time steering or retrieval. Researchers evaluate these methods on instruction-following, preference, jailbreak, factuality, multilingual, and multi-turn tasks, while also developing ways to represent disagreement among users and communities. A central challenge is that gains in safety or preference alignment may reduce usefulness, diversity, calibration, or capability, and desired behavior may not generalize to new prompts, groups, or deployment settings.

Latest papers

613 papers58 papers in the last four weeks, up 12% on the four weeks before. 0.3% of all new papers.

Jan 8, 2026

Identifying and Mitigating Bottlenecks in Role-Playing Agents: A Systematic Study of Disentangling Character Profile Axes

While Large Language Model (LLM) role-playing agents have advanced rapidly, it remains unclear which profile elements genuinely drive role-playing quality. To bridge this gap, we introduce a systematic diagnostic framework that disentangles the impact of character profiles along three axes: Familiarity (Known vs. Unknown), Structure (Structured vs. Unstructured), and Disposition (Moral vs. Immoral). Utilizing a unified hierarchical schema (5 dimensions, 28 fields), we construct a controlled dataset of 211 personas and evaluate five LLMs on both single- and multi-turn interactions. Our results reveal a striking asymmetry: Familiarity and Structure show negligible impact, while Disposition produces large, consistent performance degradation for immoral characters across all conditions. Further analyses suggest that the Moral--Immoral gap is amplified by post-SFT alignment, and that this degradation varies substantially across profile attributes. To mitigate this bottleneck, we propose Field-Aware Contrastive Decoding (FACD), a training-free strategy that amplifies suppressed disposition-sensitive signals, significantly closing the performance gap without sacrificing moral-character performance.

Yonghyun Jun, Junhyuk Choi, Jeonghyun Park +3

Chung-Ang University, Seoul, Korea

LLM Alignment Personality Modeling in Language Models LLM Agent Evaluation Conversational Agents

Jan 7, 2026

Safety Is Not Universal: The Selective Safety Trap in LLM Alignment

Current safety evaluations of large language models (LLMs) create a dangerous illusion of universal protection by aggregating harms under generic categories such as "Identity Hate", obscuring vulnerabilities toward specific populations. In this work, we expose the Selective Safety Trap: a systemic failure mode where models robustly defend specific populations while leaving underrepresented communities highly vulnerable to identical adversarial attacks. To systematically audit this phenomenon, we introduce MiJaBench, a bilingual (English-Portuguese) adversarial benchmark comprising 43,961 controlled jailbreaking prompts across 16 minority groups. By evaluating 14 state-of-the-art LLMs on MiJaBench, we curate 615,454 prompt-response pairs that compose MiJaBench-Align, revealing that safety alignment is not a uniform semantic capability but a demographic hierarchy, with defense rates fluctuating by up to 42% within the same model solely based on the target group. This disparity persists across architectures and languages and is amplified by scaling, indicating that current alignment methods learn group-specific safeguards rather than a generalized notion of harm. Through targeted direct preference optimization (DPO) on a 1B-parameter baseline, we achieve strong zero-shot safety generalizations to entirely unseen demographics and complex attack strategies. We release all datasets and scripts to provide the community with a concrete pathway toward equitable, transferable safety alignment.

Iago Alves Brito, Walcy Santos Rezende Rios, Julia Soares Dollis +2

Advanced Knowledge Center for Immersive Technologies (AKCIT) Federal University of Goiás (UFG)

LLM Safety Benchmarks LLM Alignment Language Model Safety Evaluation Adversarial Attacks on LLMs LLM Jailbreak Attacks Demographic Bias in Language Models

Jan 7, 2026

Value-Action Alignment in Large Language Models under Privacy-Prosocial Conflict

Large language models (LLMs) are increasingly used to simulate decision-making tasks involving personal data sharing, where privacy concerns and prosocial motivations can push choices in opposite directions. Existing evaluations often measure privacy-related attitudes or sharing intentions in isolation, which makes it difficult to determine whether a model's expressed values jointly predict its downstream data-sharing actions as in real human behaviors. We introduce a context-based assessment protocol that sequentially administers standardized questionnaires for privacy attitudes, prosocialness, and acceptance of data sharing within a bounded, history-carrying session. To evaluate value-action alignments under competing attitudes, we use multi-group structural equation modeling (MGSEM) to identify relations from privacy concerns and prosocialness to data sharing. We propose Value-Action Alignment Rate (VAAR), a human-referenced directional agreement metric that aggregates path-level evidence for expected signs. Across multiple LLMs, we observe stable but model-specific Privacy-PSA-AoDS profiles, and substantial heterogeneity in value-action alignment.

Guanyu Chen, Chenxiao Yu, Xiyang Hu

Arizona State University · University of Southern California

LLM Evaluation LLM Alignment

Jan 6, 2026

Metaphors are a Source of Cross-Domain Misalignment of Large Reasoning Models

Earlier research has shown that metaphors influence human decision-making, raising the question of whether metaphors also influence large language models (LLMs)' reasoning pathways, given that their training data contain a large number of metaphors. In this work, we investigate the problem in the scope of the emergent misalignment problem, where LLMs can generalize patterns learned from misaligned content in one domain to another domain. We find strong evidence that metaphors in training data contribute to cross-domain misalignment in LLMs' reasoning outputs. With metaphor-based interventions during continued pre-training and fine-tuning for inducing misalignment, models exhibit significantly different degrees of emergent cross-domain misalignment. We also observe similar effects in re-alignment settings. As we further investigate this phenomenon, we find that metaphors are linked to the activation of latent features in large reasoning models. By monitoring these latent features, we design a detector that predicts misaligned content with high accuracy.

Zhibo Hu, Chen Wang, Yanfeng Shu +2

The University of New South Wales · CSIRO Data61

LLM Alignment Emergent Misalignment in Language Models Large Reasoning Models

Jan 6, 2026

HAL: Inducing Human-likeness in LLMs with Alignment

Aligning language models to qualitative behavioral traits, such as human-likeness, remains difficult because they are hard to define, measure, and optimize. As a result, improvements in human-like behavior are largely driven by scale or broad supervised training, rather than targeted alignment. We introduce Human Aligning LLMs (HAL), a framework for aligning language models to conversational human-likeness using an interpretable, data-driven reward. HAL derives explicit conversational traits from contrastive dialogue data, combines them into a compact scalar score, and uses this score as a transparent reward signal for alignment with standard preference optimization methods. Using this approach, we align models of varying sizes without affecting their overall performance. In large-scale Chatbot Arena-style human evaluations, a model aligned with HAL is more frequently perceived as human-like in conversation. Because HAL operates over explicit, interpretable traits, it enables inspection of alignment behavior and diagnosis of unintended effects. More broadly, HAL demonstrates how soft, qualitative properties of language--previously outside the scope for alignment--can be made measurable and aligned in an interpretable and explainable way.

Masum Hasan, Junjie Zhao, Ehsan Hoque

University of Rochester

Reward Modeling LLM Alignment Human Preference Modeling

Jan 6, 2026

ECLIPTICA -- A Framework for Switchable LLM Alignment via CITA - Contrastive Instruction-Tuned Alignment

Alignment in large language models (LLMs) is still largely static: after training, the policy is frozen. DPO, GRPO methods typically imprint one behavior into the weights, leaving little runtime control beyond prompt hacks or expensive re-alignment. We introduce ECLIPTICA, which treats alignment as instruction-driven and runtime-controllable: natural-language alignment instructions act as an explicit behavioral contract (stance, refusal boundary, verbosity) that modulates behavior on the fly under evolving safety requirements, user roles, and governance constraints. We introduce CITA (Contrastive Instruction-Tuned Alignment), combining SFT with contrastive preference optimization under an explicit geometric anchor to a reference model. This yields a stable Riemannian chart and keeps instruction updates within a shared neighborhood, so regimes stay nearby and traversable for reliable switching. To isolate policy switching from ordinary instruction following, we release the ECLIPTICA benchmark: 3000 controlled cases (300 prompts x 10 instruction types) where the user request is fixed and only the alignment instruction changes. On Llama-3.1-8B across five suites (ECLIPTICA, TruthfulQA, Conditional Safety, Length Control, LITMUS), CITA reaches 86.7% instruction-alignment efficiency, beating DPO (56.1%), GRPO (36.1%), and PPO (20.4%).

Kapil Wanaskar, Gaytri Jena, Vinija Jain +2

San José de State University, USA · Pragya Lab, BITS Pilani Goa, India · Google, USA +1

LLM Alignment Inference-Time Language Model Alignment Preference Optimization

Dec 18, 2025

AlignMerge - Alignment-Preserving Large Language Model Merging via Fisher-Guided Geometric Constraints

Merging large language models (LLMs) is a practical way to compose capabilities from multiple fine-tuned checkpoints without retraining. Yet standard schemes (linear weight soups, task vectors, and Fisher-weighted averaging) can preserve loss while quietly destroying alignment. We argue that merging is not a numerical trick but a geometry-constrained operation around an already-aligned anchor: fusion must be steered to respect safety geometry, not validated post hoc. We introduce AlignMerge, a geometry-aware merging framework that makes alignment an explicit invariant. In a local Fisher chart around an instruction-tuned base, we estimate an alignment subspace with projector P_A and optimize: L_AlignMerge = L_geo + lambda_align * L_align + lambda_bud * L_bud, where L_geo keeps the merge close to its experts in Fisher-Rao geometry, L_align penalizes motion along alignment-sensitive directions, and L_bud enforces a soft alignment budget. As the alignment functional we use the decoding-invariant Alignment Quality Index (AQI), a latent-space criterion that captures how cleanly aligned and misaligned behaviors separate in representation space. Across five model families (LLaMA-3 8B, Mistral 7B, Qwen 2, Phi-3.5, Gemma 2), merging safety anchors with task experts, AlignMerge improves alignment metrics (AQI, toxicity, LLM-judge alignment) while matching or exceeding the best expert on instruction-following, reasoning, and helpfulness. It also exhibits smaller alignment-subspace drift and fewer budget violations than Fisher soups, TIES, SafeMerge, and MergeAlign. These results make alignment-preserving merging a first-class design goal and suggest a path to geometry-aware composition of future foundation models.

Aniruddha Roy, Jyoti Patel, Aman Chadha +2

HCL · Evalueserve · Apple (USA) +2

Language Model Merging LLM Alignment LLM Safety Alignment Latent Space Alignment

Dec 12, 2025

Shapley-based Data Valuation for LLM Alignment via Sequential Preference Optimization

Data valuation is a natural framework for understanding which preference datasets matter most when aligning a Large Language Model (LLM) using multiple sources. The standard game-theoretic approach assigns each dataset a contribution score via the Shapley value. In practice, however, Shapley-based valuation is computationally prohibitive because it requires fine-tuning a separate model for every possible coalition of preference datasets, i.e., an exponential number of alignments. We address this challenge for a broad family of preference-optimization objectives, including DPO and IPO, that learn directly from log-policy ratios with respect to a reference policy. We introduce Sequential Preference Optimization, an offline procedure that applies existing preference optimization methods sequentially, source by source, updating the current policy after each dataset. Under exact optimization, this procedure yields an additive composition rule in reward space and an equivalent arithmetic composition rule in policy space. This observation enables an efficient approximation of the Shapley value: we train one model per preference dataset and reconstruct coalition policies at inference time from the singleton models, reducing the required alignments from exponential to linear in the number of sources. Leveraging this property, we compute Shapley values for several real-world preference datasets and reveal how each source drives model alignment.

Mélissa Tamine, Otmane Sakhi, Benjamin Heymann +2

Criteo AI lab, FairPlay joint team, France · CREST, ENSAE, Institut Polytechnique de Paris · Criteo AI Lab, France +1

LLM Alignment Shapley Value Attribution Preference Optimization Preference Alignment Data Valuation

Dec 3, 2025

Overcoming State Inertia: Minimally Invasive Temporal Alignment for Evolving Contexts

Long-context dialogue systems suffer from state inertia, where models over-attend to history and fail to adapt to evolving intents. We demonstrate that standard alignment methods like DPO and even recent long-context optimization techniques struggle to resolve this without incurring a severe contextual alignment tax--a substantial perplexity surge caused by disrupting pre-trained priors. To address this, we propose DZ-TiDPO, a minimally invasive framework that synergizes conflict-aware optimization (during training) with a structural temporal attention bias. This design effectively decouples state updating from general linguistic modeling. Experiments on Multi-Session Chat and our new Inertia Challenge (IC-Bench) show DZ-TiDPO preserves structural coherence while resolving inter-turn conflicts. Crucially, our framework supports dual inference strategies: a negligible-latency static mode for general robustness and a precision-focused dynamic mode for micro-semantic conflicts. Furthermore, our scaling analysis reveals a capacity-stability trade-off, confirming that highly capable mid-sized models (7B) can efficiently internalize temporal alignment. Code and data are available at: https://github.com/lyj20071013/DZ-TiDPO.

Yijun Liao

LLM Alignment Direct Preference Optimization Long-Context Modeling Temporal Attention

Nov 26, 2025

Aligning LLMs with Biomedical Knowledge using Balanced Fine-Tuning

Engineering LLMs to accelerate life sciences research requires a robust alignment with biomedical knowledge. We observe that biomedical text exhibits a fundamentally different uncertainty structure from general text: dense low-confidence runs encode epistemic knowledge gaps (dense causal chains, rare entities) rather than the sparse aleatoric stylistic variation typical of general text. Based on this discovery, we propose Balanced Fine-Tuning (BFT), a dual-scale post-training method that combines group-normalized token reweighting with sequence-level reallocation toward knowledge-dense samples exhibiting dense epistemic uncertainty. Across medical evaluation, biological reasoning, sparse-reward RL, and biological representation tasks, BFT provides more consistent gains than SFT and DFT under a shared training setup. When replacing the default closed-source backbones in GeneAgent (GPT-4o) and VCWorld (Gemini-2.5-Flash), the BFT-aligned 70B model delivers stronger performance across biological process reasoning and chemical perturbation prediction. Critically, all BFT variants further improve after subsequent GRPO with sparse rewards, while SFT and DFT degrade, suggesting that epistemic-aware post-training provides a more robust policy initialization. Beyond text generation, BFT-aligned LLMs produce more accurate and professional biomedical profile texts; after encoding these profiles with a text embedding model, the resulting representations support gene-level, cell-level, and perturbation-response tasks, suggesting that BFT-enhanced generation can facilitate biological representation and, in turn, broader biomedical downstream tasks.

Zhenchao Tang, Fang Wang, Haohuai He +13

Tencent AI for Life Sciences Lab, Shenzhen, China. · Sun Yat-sen University, Shenzhen, China. · The Hong Kong Polytechnic University, Hong Kong SAR, China. +2

LLM Alignment LLM Fine-Tuning LLM Uncertainty Estimation LLM Post-Training Medical LLMs

Nov 22, 2025

Alignment Faking - the Train -> Deploy Asymmetry: Through a Game-Theoretic Lens with Bayesian-Stackelberg Equilibria

Alignment faking is a form of strategic deception in AI in which models selectively comply with training objectives when they infer that they are in training, while preserving different behavior outside training. The phenomenon was first documented for Claude 3 Opus and later examined across additional large language models. In these setups, the word "training" refers to simulated training via prompts without parameter updates, so the observed effects are context conditioned shifts in behavior rather than preference learning. We study the phenomenon using an evaluation framework that compares preference optimization methods (BCO, DPO, KTO, and GRPO) across 15 models from four model families, measured along three axes: safety, harmlessness, and helpfulness. Our goal is to identify what causes alignment faking and when it occurs.

Kartik Garg, Shourya Mishra, Kartikeya Sinha +8

Pragya Lab, BITS Pilani Goa, India · Apple, USA · Google, USA

Deception in Language Models LLM Alignment

Oct 30, 2025

Value Drifts: Tracing Value Alignment During LLM Post-Training

As LLMs occupy an increasingly important role in society, they are more and more confronted with questions that require them not only to draw on their general knowledge but also to align with certain human value systems. Therefore, studying the alignment of LLMs with human values has become a crucial field of inquiry. Prior work, however, mostly focuses on evaluating the alignment of fully trained models, overlooking the training dynamics by which models learn to express human values. In this work, we investigate how and at which stage value alignment arises during the course of a model's post-training. Our analysis disentangles the effects of post-training algorithms and datasets, measuring both the magnitude and time of value drifts during training. Experimenting with Llama-3 and Qwen-3 models of different sizes and popular supervised fine-tuning (SFT) and preference optimization datasets and algorithms, we find that the SFT phase generally establishes a model's values, and subsequent preference optimization rarely re-aligns these values. Furthermore, using a synthetic preference dataset that enables controlled manipulation of values, we find that different preference optimization algorithms lead to different value alignment outcomes, even when preference data is held constant. Our findings provide actionable insights into how values are learned during post-training and help to inform data curation, as well as the selection of models and algorithms for preference optimization to improve model alignment to human values.

Mehar Bhatia, Shravan Nayak, Gaurav Kamath +4

Mila - Quebec AI Institute · McGill University · Université de Montréal +4

LLM Alignment Preference Optimization LLM Post-Training

Oct 28, 2025

OpenReward: Learning to Reward Long-form Agentic Tasks via Reinforcement Learning

Reward models (RMs) have become essential for aligning large language models (LLMs), serving as scalable proxies for human evaluation in both training and inference. However, existing RMs struggle on knowledge-intensive and long-form tasks, where evaluating correctness requires grounding beyond the model's internal knowledge. This limitation hinders them from reliably discriminating subtle quality differences, especially when external evidence is necessary. To address this, we introduce OpenRM, a tool-augmented long-form reward model that systematically judges open-ended responses by invoking external tools to gather relevant evidence. We train OpenRM with Group Relative Policy Optimization (GRPO) on over 27K synthesized pairwise examples generated through a controllable data synthesis framework. The training objective jointly supervises intermediate tool usage and final outcome accuracy, incentivizing our reward model to learn effective evidence-based judgment strategies. Extensive experiments on three newly-collected datasets and two widely-used benchmarks demonstrate that OpenRM substantially outperforms existing reward modeling approaches. As a further step, we integrate OpenRM into both inference-time response selection and training-time data selection. This yields consistent gains in downstream LLM alignment tasks, highlighting the potential of tool-augmented reward models for scaling reliable long-form evaluation.

Ziyou Hu, Zhengliang Shi, Minghang Zhu +5

Leiden University, Leiden, Netherland · Shandong University, Qingdao, China · Tsinghua University, Beijing, China

Reward Modeling RL for Language Models LLM Alignment

Oct 20, 2025

Navigating the Alignment-Calibration Trade-off: A Pareto-Superior Frontier via Model Merging

The "alignment tax" of post-training is typically framed as a drop in task accuracy. We show it also involves a severe loss of calibration, making models overconfident, less reliable, and model outputs less diverse. We show that this trade-off can be navigated effectively via a simple post-hoc intervention: interpolating between a model's weights before and after alignment. Crucially, this is not a strict trade-off. We find that the process consistently reveals Pareto-optimal interpolations - models that improve accuracy beyond both parents while substantially recovering the calibration lost during alignment. Our work demonstrates that simple model merging provides a computationally efficient method for mitigating the full scope of the alignment tax, yielding models that are more capable and more reliable.

Tiancheng Hu, Benjamin Minixhofer, Nigel Collier

University of Cambridge

LLM Alignment Model Calibration Multi-Objective Language Model Alignment Language Model Merging

Oct 15, 2025

The Alignment Veto: How Safety Training Suppresses Cultural Knowledge in LLMs

What happens inside a language model when alignment training conflicts with a cultural value it encodes? Across 16 MENA countries, 26 models, and 1.53M human survey responses, we show the answer is suppression, not erasure: at the moment of refusal, a model's internal logit distribution correlates with human survey data more strongly than its freely generated answers. We call this the alignment veto. We distinguish suppression failures (accurate internal distributions blocked at output) from representational bias failures (the encoding itself diverges from human values), and show the two require different interventions. The gate is inequitable: the safety tax reaches 37.6%, with a 19.8% alignment-quality gap between best- and worst-served nations, and native-language prompting widens rather than closes it. Sparse autoencoder analysis corroborated by comparisons across alignment stages identifies a candidate DPO-stage feature mediating suppression in Tulu-3-8B. The gate is real, its costs are unequal, and deciding what it should protect is not a technical question.

Pardis Sadat Zahraei, Gokhan Tur, Dilek Hakkani-Tür +1

University of Illinois Urbana-Champaign (UIUC) · Qatar Computing Research Institute (QCRI), Hamad Bin Khalifa University

LLM Alignment Refusal Behavior in Language Models Cultural Knowledge in Language Models Cross-Cultural Language Model Evaluation

Oct 14, 2025

Hierarchical Alignment: Block-Level LoRA-DPO for Preference Tuning in Large Language Models

Preference tuning typically distributes adapter updates across every Transformer layer. We study how the placement of a fixed LoRA budget affects Direct Preference Optimization (DPO). Hierarchical Alignment partitions decoder depth into Local, Intermediate, and Global regions and restricts LoRA updates to a chosen contiguous block. Across Qwen and Llama models, three-seed sliding-window sweeps reveal broad low-loss basins around intermediate depth. Under exact trainableparameter matching, Middle achieves lower held-out DPO loss than an all-depth allocation on Qwen2.5-7B and Llama-3.1-8B. Generationlevel evidence is more variable. LLM judges and a blinded human study leave Selected versus All unresolved on Qwen; on Llama, two pairwise judges favor All while the human interval includes zero. Trained-adapter interventions show that learned depth components affect the optimization objective; removing the Llama Selected window also lowers generation quality. Across Qwen2.5 at 3B, 7B, and 14B, Middle versus All remains unresolved under Qwen3-Max; at 3B and 14B, both improve over Base. In our implementation, Middle also uses less peak memory and takes shorter steps than a parameter-matched all-depth allocation. Together, the results support intermediate depth as a practical starting point for fixed-budget preference tuning, though the best window and its behavioral value still need per-setting validation.

Yishen Chen, Qi Dong, Yukun Zhang

Fudan University Shanghai, China · The Chinese University of Hong Kong Hong Kong, China

LLM Alignment Preference Alignment Parameter-Efficient Fine-Tuning Direct Preference Optimization Low-Rank Adaptation

Oct 11, 2025

Interaction Protocol Shapes Moral Judgment in Multi-Agent Debate

As large language models (LLMs) are increasingly deployed in sensitive everyday contexts -- offering personal advice, mental health support, and moral guidance -- understanding their behavior in navigating complex moral reasoning is essential. Most evaluations study this sociotechnical alignment through single-turn prompts, but it is unclear if these findings extend to multi-turn settings, and even less clear how they depend on the interaction protocols used to coordinate agentic systems. We address this gap using LLM debate to examine deliberative dynamics and value alignment in multi-turn settings by prompting subsets of three models (GPT-4.1, Claude 3.7 Sonnet, and Gemini 2.0 Flash) to collectively assign blame in 1,000 everyday dilemmas from Reddit's ``Am I the Asshole'' community. To test order effects and assess verdict revision, we use both synchronous (parallel responses) and round-robin (sequential responses) deliberation structures, mirroring how multi-agent systems are increasingly orchestrated in practice. Our findings show striking behavioral differences. In the synchronous setting, GPT-4.1 showed strong inertia (0.6-3.1% revision rates) while Claude 3.7 Sonnet and Gemini 2.0 Flash were far more flexible (28-41% revision rates). Value patterns also diverged: GPT-4.1 emphasized personal autonomy and direct communication (relative to its deliberation partners), while Claude 3.7 Sonnet and Gemini 2.0 Flash prioritized empathetic dialogue. We further find that deliberation format had a strong impact on model behavior: GPT-4.1 and Gemini 2.0 Flash stood out as highly conforming relative to Claude 3.7 Sonnet, with their verdict behavior strongly shaped by order effects. We provide additional results on open-source models (DeepSeek-V3.2 and Llama 3.1).

Pratik S. Sachdeva, Tom van Nuenen

D-Lab, University of California, Berkeley

LLM Decision-Making Moral Reasoning in Language Models Multi-Agent Debate LLM Alignment Multi-Agent Systems Moral Reasoning+1

Oct 7, 2025

Moloch's Bargain: Emergent Misalignment When LLMs Compete for Audiences

Large language models (LLMs) are increasingly shaping how information is created and disseminated, from companies using them to craft persuasive advertisements, to election campaigns optimizing messaging to gain votes, to social media influencers boosting engagement. These settings are inherently competitive, with sellers, candidates, and influencers vying for audience approval, yet it remains poorly understood how competitive feedback loops influence LLM behavior. We show that optimizing LLMs for competitive success can inadvertently drive misalignment. Using simulated environments across these scenarios, we find that, 6.3% increase in sales is accompanied by a 14.0% rise in deceptive marketing; in elections, a 4.9% gain in vote share coincides with 22.3% more disinformation and 12.5% more populist rhetoric; and on social media, a 7.5% engagement boost comes with 188.6% more disinformation and a 16.3% increase in promotion of harmful behaviors. We call this phenomenon Moloch's Bargain for AI--competitive success achieved at the cost of alignment. These misaligned behaviors emerge even when models are explicitly instructed to remain truthful and grounded, revealing the fragility of current alignment safeguards. Our findings highlight how market-driven optimization pressures can systematically erode alignment, creating a race to the bottom, and suggest that safe deployment of AI systems will require stronger governance and carefully designed incentives to prevent competitive dynamics from undermining societal trust.

Batu El, James Zou

Stanford University

LLM Alignment LLM Safety

Sep 27, 2025

TagPR: Tag-Guided Process Supervision for Personalization Reasoning in Large Language Models

Recent advancements have endowed Large Language Models with impressive general reasoning capabilities. However, these reasoning models often perform worse than non-reasoning models on personalization tasks. While some methods use outcome-based RL to improve personalization reasoning, they fail to supervise the reasoning process. As a result, models may reach correct answers through flawed reasoning chains, limiting further improvement. To address this, we propose TagPR, a novel framework that adds semantic tags to the reasoning process for step-by-step guidance. TagPR first automatically generates a structured, tagged dataset for Supervised Fine-Tuning. It then employs a multi-stage RL process guided by a composite reward signal, which integrates tag-based process supervision with a novel Personalization Reward Model with User Embeddings to achieve fine-grained alignment with user-specific logic. Extensive experiments on public LaMP, LongLaMP, PGraphRAG, and a self-constructed dataset demonstrate that our approach achieves state-of-the-art results, delivering an average improvement of 32.65% over the base model across all LaMP benchmark tasks. Our work demonstrates that tag-guided process supervision is an effective approach for personalization reasoning.

Song Jin, Juntian Zhang, Ruyu Lyu +7

Gaoling School of Artificial Intelligence, Renmin University of China · Meituan · Wuhan University

LLM Alignment LLM Personalization RL for Language Model Reasoning Process Supervision

Aug 15, 2025

From Value Conditioning to Behavioral Shift: Lightweight Value Alignment of LLMs

Large language models exhibit value-related preferences in their outputs, yet steering such behavior often requires large training datasets or paired preference annotations. In this work, we investigate a lightweight alternative: can fine-tuning on value survey responses induce measurable shifts in value-conditioned model outputs and downstream behavior? We first estimate baseline value profiles for several open-source LLMs by asking them to rate value-related descriptions organized around the ten Schwartz values. We then fine-tune models on survey-style supervision that selectively changes the target value ratings while keeping non-target values close to their baseline responses. We evaluate the resulting shifts in two settings. First, we measure whether the intervention changes responses on held-out survey questions. Second, we test whether these survey-induced shifts transfer to out-of-domain behavioral proxies, including a contextual moral-judgment benchmark constructed from Reddit AITA posts and a text-based interactive decision-making benchmark. Our results show that survey-based supervision can change held-out survey responses and produce downstream shifts in model response distributions on proxy behavioral evaluations. However, this transfer is strongly model- and value-dependent, and is shaped by how value descriptions are framed in the supervision data. These findings suggest that value surveys can serve not only as diagnostic tools, but also as lightweight supervision for studying behavioral value conditioning in LLMs.

Shangrui Nie, Florian Mai, David Kaczér +3

Conversational AI and Social Analytics (CAISA) Lab, University of Bonn, Germany · McMaster University, Canada · Department of Computer Science, University of Sheffield, United Kingdom

LLM Alignment Supervised Fine-Tuning

Jul 6, 2025

ESSA: Evolutionary Strategies for Scalable Alignment

Online alignment of large language models (LLMs) is dominated by reinforcement learning from human feedback (RLHF) with gradient-based optimizers such as PPO or GRPO. While effective, these pipelines require backpropagation through long rollouts, gradient synchronization across devices, and careful hyperparameter tuning, all of which become increasingly costly at scale. We present ESSA (Evolutionary Strategies for Scalable Alignment), a gradient-free online alignment stage that follows supervised fine-tuning (SFT) and replaces the gradient loop with inference-only black-box search. ESSA optimizes only the singular values of low-rank adaptation (LoRA) factors after a short SFT warm-start, restricting the search to a compact, task-aligned subspace where evolutionary search is practical even for 72B-parameter models. Because the loop is inference-only, ESSA is compatible with INT4/INT8 weight quantization and reduces inter-GPU communication to a few bytes per iteration. Across instruction following (IFEval), preference-based assistant tuning (HelpSteer2, HH-RLHF), and mathematical reasoning (GSM8K, MATH500), ESSA matches or exceeds LoRA-GRPO in the reported LoRA comparisons; on GSM8K it also outperforms Online DPO and PPO, while remaining competitive with both methods on IFEval. At scale, ESSA reaches a fixed MATH500 accuracy on Qwen2.5-32B/PRM800K up to 7.8x faster than LoRA-GRPO on 128 GPUs.

Daria Korotyshova, Boris Shaposhnikov, Alexey Malakhov +7

T-Tech · Yandex

LLM Alignment Evolutionary Optimization Evolution Strategies

Jul 5, 2025

A Technical Survey of Reinforcement Learning Techniques for Large Language Models

This survey offers a comprehensive foundation on the integration of RL with language models, highlighting prominent algorithms such as Proximal Policy Optimization (PPO), Q-Learning, and Actor-Critic methods. Additionally, it provides an extensive technical overview of RL techniques specifically tailored for LLMs, including foundational methods like Reinforcement Learning from Human Feedback (RLHF) and AI Feedback (RLAIF), as well as advanced strategies such as Direct Preference Optimization (DPO) and Group Relative Policy Optimization (GRPO). We systematically analyze their applications across domains, i.e., from code generation to tool-augmented reasoning. Crucially, we move beyond descriptive categorization to provide a rigorous algorithmic analysis of failure modes, mathematically framing the structural bottlenecks and stability trade-offs inherent in policy optimization. We also present a comparative taxonomy based on reward modeling, feedback mechanisms, and optimization strategies. Our evaluation highlights key trends. RLHF remains dominant for alignment, and outcome-based RL such as Reinforcement Learning with Verifiable Rewards (RLVR) significantly improves stepwise reasoning. However, persistent challenges such as reward hacking, computational costs, and scalable feedback collection underscore the need for continued innovation. We also explicate the causal factors behind recent benchmark performances, distinguishing between gains derived from architectural scaling versus those stemming from specific optimization objectives. We further discuss emerging directions, including hybrid RL algorithms, verifier-guided training, and multi-objective alignment frameworks. This survey serves as a roadmap for researchers advancing RL-driven LLM development, balancing capability enhancement with safety and scalability.

Saksham Sahai Srivastava, Vaneet Aggarwal

University of Georgia, Athens, GA, USA · Purdue University, West Lafayette, IN, USA

RL for Language Models LLM Alignment Direct Preference Optimization RL from Human Feedback

Jun 30, 2025

Logit-Gap Steering: A Forward-Pass Diagnostic for Alignment Robustness

RLHF-style alignment trains language models to refuse unsafe requests, but how much operational margin does this refusal rest on? We introduce the refusal-affirmation logit gap: the difference between the top refusal-token logit and the top affirmative-token logit at the first decoding step. This single scalar quantifies the per-prompt safety margin that alignment provides. Empirically, alignment widens the gap on 97.5-99.8% of toxic prompts across three model families, and median gap closure co-varies with True-ASR ranking across suffix strategies (an internal consistency check, since our method optimises gap closure). To validate the metric's practical significance, we present logit-gap steering, a gradient-free, forward-pass-only method that discovers short in-distribution suffixes (

<

10 tokens per component) whose cumulative effect closes the gap. The method requires

{\approx}26{,}000

forward-pass equivalents per family (

{\approx}2

~min on one A100),

{\approx}125\times

less than a single GCG search. Suffixes discovered on 0.5B--2B models transfer without modification to 72B within family. An 8-suffix ensemble reaches 38-96% True ASR across 13 models on AdvBench and HarmBench, with most suffixes having

10^{3}

-

10^{4}\times

lower perplexity than GCG-meaning published perplexity-filter defenses that collapse GCG (64.7%

\to

1.0%) leave our suffixes nearly intact (76.9%

\to

76.0%). These results demonstrate that current alignment margins, while consistently present, can be thin and efficiently measurable, and that defense strategies must account for in-distribution suffixes.

Tung-Ling Li, Hongliang Liu

Palo Alto Networks

LLM Alignment Language Model Safety Evaluation Adversarial Attacks on LLMs Refusal Behavior in Language Models

Jun 3, 2025

KARE-RAG: Knowledge-Aware Refinement and Enhancement for RAG

Retrieval-Augmented Generation (RAG) equips large language models with external knowledge and is central to knowledge-intensive tasks. As RAG systems enter real-world use, generators must reliably leverage retrieved evidence. Recent fine-tuning methods improve adaptation to RAG scenarios, but optimization remains challenging because retrieval may return incomplete, fragmented, noisy, or conflicting contexts. Complex tasks further require fine-grained evidence dependencies. These challenges make high-quality supervision costly and limit generalization. We present KARE-RAG (Knowledge-Aware Refinement and Enhancement for RAG), a training-time scaffolded alignment framework. It uses structured knowledge representations as temporary scaffolds to expose evidence organization, support localized factual refinement, and construct fine-grained preference pairs. In our main implementation, an expert LLM refines a lightweight graph-structured evidence sketch. The generator is optimized with token-weighted Dense Direct Preference Optimization (DDPO), which focuses learning on edited scaffold regions. Scaffolds are used only for data construction and training supervision. At inference time, the model runs as standard Vanilla RAG without graph construction, extra retrieval, or latency overhead. Experiments show that KARE improves transfer across the evaluated QA and relation extraction datasets with limited training data, while leaving general capabilities largely unchanged. KARE can also complement existing RAG training objectives as an additional alignment stage.

Yongjian Li, HaoCheng Chu, Yukun Yan +7

School of Biomedical Engineering, Tsinghua Medicine, Tsinghua University · 2Tsinghua Laboratory of Brain and Intelligence · School of Informatics, Xiamen University +6

Retrieval-Augmented Generation LLM Alignment

Jun 2, 2025

C-VARC: A Large-Scale Chinese Value Rule Corpus for Value Alignment of Large Language Models

Aligning LLMs with Chinese values requires culturally grounded resources that connect abstract values to concrete behavioral guidance. We introduce C-VARC, a Chinese Value Rule Corpus containing 257,609 Rules of Thumb (RoTs) organized into three dimensions, 12 core values, and 50 derived values. Among them, 36,000 rules underwent independent human review by two annotators (Cohen's kappa>0.85), with disagreements adjudicated by a third. C-VARC supports reusable rule-driven pipelines for both value evaluation and model alignment. Compared with abstract-value guidance, RoT guidance produces greater semantic and character-level n-gram diversity. Across five non-political sensitive themes, the seven-model average selection rate for C-VARC-consistent options is at least 72.0%, while five Chinese annotators show overall rates of at least 87.5%. Conflicting RoT pairs further enable the construction of 404,505 moral dilemmas, with 7,191 of 10,998 evaluated cases eliciting cross-model disagreement. For alignment, C-VARC-Query consistently outperforms matched Value-only supervision on CMoralEval across both Qwen3-8B and Llama-3.1-8B-Instruct, with average gains of 2.97 and 12.82 percentage points, respectively, and positive paired gains across all three seeds for both backbones. It also improves CValues and C-Eval over Value-only supervision on both backbones. C-VARC thus provides a scalable Chinese normative resource for constructing both value-evaluation and alignment supervision.

Ping Wu, Guobin Shen, Dongcheng Zhao +6

BrainCog Lab, Institute of Automation, Chinese Academy of Sciences, Beijing, China · The School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China · Beijing Key Laboratory of Safe AI and Superalignment, Beijing, China +2

Value Alignment LLM Alignment Moral Reasoning in Language Models LLM Evaluation Cultural Knowledge in Language Models

Jun 1, 2025

Doubly Robust Alignment for Large Language Models

This paper studies reinforcement learning from human feedback (RLHF) for aligning large language models with human preferences. While RLHF has demonstrated promising results, many algorithms are highly sensitive to misspecifications in the underlying preference model (e.g., the Bradley-Terry model), the reference policy, or the reward function, resulting in undesirable fine-tuning. To address model misspecification, we propose a doubly robust preference optimization algorithm that remains consistent when either the preference model or the reference policy is correctly specified (without requiring both). Our proposal demonstrates superior and more robust performance than state-of-the-art algorithms, both in theory and in practice. The code is available at https://github.com/DRPO4LLM/DRPO4LLM

Erhan Xu, Kai Ye, Hongyi Zhou +3

Department of Statistics, LSE, London, UK · Department of Mathematics, Tsinghua University, Beijing, China · School of Design, LCC, UAL, London, UK +1

LLM Alignment RL from Human Feedback Preference Optimization Preference-Based RL

May 30, 2025

Aligning Language Models with Observational Data: Opportunities and Risks from a Causal Perspective

Large language models are being widely used across industries to generate text that contributes directly to key performance metrics, such as medication adherence in patient messaging and conversion rates in content generation. Pretrained models, however, often fall short when it comes to aligning with human preferences or optimizing for business objectives. As a result, fine-tuning with good-quality labeled data is essential to guide models to generate content that achieves better results. Controlled experiments, like A/B tests, can provide such data, but they are often expensive and come with significant engineering, logistical, and ethical challenges. Meanwhile, companies have access to a vast amount of historical (observational) data that remains underutilized. In this work, we study the challenges and opportunities of fine-tuning LLMs using observational data. We show that while observational outcomes can provide valuable supervision, directly fine-tuning models on such data can lead them to learn spurious correlations. We present empirical evidence of this issue using various real-world datasets and propose DeconfoundLM, a method that explicitly removes the effect of known confounders from reward signals. In simulation experiments, DeconfoundLM more accurately recovers causal relationships and mitigates failure modes of methods that assume counterfactual invariance, achieving over 16% higher objective score than ODIN and other baselines, when entangled confounding is present. Please refer to the project page for code and related resources.

Erfan Loghmani

Foster School of Business University of Washington Seattle, WA 98195

LLM Alignment LLM Fine-Tuning Causal Inference

May 20, 2025

InfiFPO: Implicit Model Fusion via Preference Optimization in Large Language Models

Model fusion combines multiple Large Language Models (LLMs) with different strengths into a more powerful, integrated model through lightweight training methods. Existing works on model fusion focus primarily on supervised fine-tuning (SFT), leaving preference alignment (PA) --a critical phase for enhancing LLM performance--largely unexplored. The current few fusion methods on PA phase, like WRPO, simplify the process by utilizing only response outputs from source models while discarding their probability information. To address this limitation, we propose InfiFPO, a preference optimization method for implicit model fusion. InfiFPO replaces the reference model in Direct Preference Optimization (DPO) with a fused source model that synthesizes multi-source probabilities at the sequence level, circumventing complex vocabulary alignment challenges in previous works and meanwhile maintaining the probability information. By introducing probability clipping and max-margin fusion strategies, InfiFPO enables the pivot model to align with human preferences while effectively distilling knowledge from source models. Comprehensive experiments on 11 widely-used benchmarks demonstrate that InfiFPO consistently outperforms existing model fusion and preference optimization methods. When using Phi-4 as the pivot model, InfiFPO improve its average performance from 79.95 to 83.33 on 11 benchmarks, significantly improving its capabilities in mathematics, coding, and reasoning tasks.

Yanggan Gu, Yuanyi Wang, Zhaoyi Yan +4

The Hong Kong Polytechnic University (PolyU) · Zhejiang University · PolyU-Daya Bay Technology and Innovation Research Institute

Language Model Merging LLM Alignment Direct Preference Optimization

May 19, 2025

R3: Robust Rubric-Agnostic Reward Models

Reward models are essential for aligning language model outputs with human preferences, yet existing approaches often lack both controllability and interpretability. These models are typically optimized for narrow objectives, limiting their generalizability to broader downstream tasks. Moreover, their scalar outputs are difficult to interpret without contextual reasoning. To address these limitations, we introduce R3, a novel reward modeling framework that is rubric-agnostic, generalizable across evaluation dimensions, and provides interpretable, reasoned score assignments. R3 enables more transparent and flexible evaluation of language models, supporting robust alignment with diverse human values and use cases. Our models, data, and code are available as open source at https://github.com/rubricreward/r3.

David Anugraha, Zilu Tang, Lester James V. Miranda +5

Department of Computer Science, Stanford University · Department of Computer Science, Boston University · Allen Institute for AI +4

Reward Modeling LLM Evaluation LLM Alignment Language Model Generation Evaluation

May 19, 2025

Bias Fitting to Mitigate Length Bias of Reward Model in RLHF

Reinforcement Learning from Human Feedback (RLHF) relies on reward models to align large language models with human preferences. However, RLHF often suffers from reward hacking, wherein policy learning exploits flaws in the trained reward model to maximize reward scores without genuinely aligning with human preferences. A significant example of such reward hacking is length bias, where reward models usually favor longer responses irrespective of actual response quality. Previous works on tackling length bias have notable limitations, these approaches either mitigate bias without characterizing the bias form, or simply assume a linear length-reward relation. To accurately model the intricate nature of length bias and facilitate more effective bias mitigation, we propose FiMi-RM (Bias Fitting to Mitigate Length Bias of Reward Model), a framework that autonomously learns and corrects underlying bias patterns. Our approach consists of three stages: First, we warm up by training a standard reward model which inherently contains length bias. Next, we deploy a lightweight fitting model to capture the non-linear relation between length and reward. Finally, we incorporate this learned relation into the reward model, effectively decoupling length from reward while preserving preference modeling capabilities. Experimental results demonstrate that FiMi-RM achieves a more balanced length-reward distribution. Furthermore, when applied to alignment algorithms such as Direct Preference Optimization (DPO) and Best-of-N (BoN), our debiased reward model improves length-controlled win rate and reduces verbosity without compromising its performance.

Kangwen Zhao, Jianfeng Cai, Jinhua Zhu +5

University of Science and Technology of China

Reward Modeling LLM Alignment RL from Human Feedback