Daily Papers

Inference-time steering methods offer a lightweight alternative to fine-tuning large language models (LLMs) and vision-language models (VLMs) by modifying internal activations at test time without updating model weights. However, most existing approaches rely on fixed, global intervention vectors, overlook the causal influence of individual input tokens, and fail to leverage informative gradients from the model's logits, particularly in multimodal settings where visual and textual inputs contribute unevenly. To address these limitations, we introduce GrAInS, an inference-time steering approach that operates across both language-only and vision-language models and tasks. GrAInS uses contrastive, gradient-based attribution via Integrated Gradients to identify the top-k most influential tokens, both positively and negatively attributed based on their contribution to preferred versus dispreferred outputs. These tokens are then used to construct directional steering vectors that capture semantic shifts from undesirable to desirable behavior. During inference, GrAInS adjusts hidden activations at transformer layers guided by token-level attribution signals, and normalizes activations to preserve representational scale. This enables fine-grained, interpretable, and modular control over model behavior, without retraining or auxiliary supervision. Empirically, GrAInS consistently outperforms both fine-tuning and existing steering baselines: it achieves a 13.22% accuracy gain on TruthfulQA using Llama-3.1-8B, reduces hallucination rates on MMHal-Bench from 0.624 to 0.514 with LLaVA-1.6-7B, and improves alignment win rates on SPA-VL by 8.11%, all while preserving the model's fluency and general capabilities.

Text-to-image (T2I) diffusion models have made remarkable advancements, yet aligning them with diverse preferences remains a persistent challenge. Current methods often optimize single metrics or depend on narrowly curated datasets, leading to overfitting and limited generalization across key visual quality metrics. We present BalancedDPO, a novel extension of Direct Preference Optimization (DPO) that addresses these limitations by simultaneously aligning T2I diffusion models with multiple metrics, including human preference, CLIP score, and aesthetic quality. Our key novelty lies in aggregating consensus labels from diverse metrics in the preference distribution space as compared to existing reward mixing approaches, enabling robust and scalable multi-metric alignment while maintaining the simplicity of the standard DPO pipeline that we refer to as BalancedDPO. Our evaluations on the Pick-a-Pic, PartiPrompt and HPD datasets show that BalancedDPO achieves state-of-the-art results, outperforming existing approaches across all major metrics. BalancedDPO improves the average win rates by 15%, 7.1%, and 10.3% on Pick-a-pic, PartiPrompt and HPD, respectively, from the DiffusionDPO.

Direct Preference Optimization (DPO) is an effective technique that leverages pairwise preference data (usually one chosen and rejected response pair per user prompt) to align LLMs to human preferences. In practice, multiple responses can exist for a given prompt with varying quality relative to each other. With availability of such quality ratings for multiple responses, we propose utilizing these responses to create multiple preference pairs for a given prompt. Our work focuses on systematically using the constructed multiple preference pair in DPO training via curriculum learning methodology. In particular, we order these multiple pairs of preference data from easy to hard (emulating curriculum training) according to various criteria. We show detailed comparisons of our proposed approach to the standard single-pair DPO setting. Our method, which we call Curry-DPO consistently shows increased performance gains on MTbench, Vicuna, WizardLM, and the UltraFeedback test set, highlighting its effectiveness. More specifically, Curry-DPO achieves a score of 7.43 on MT-bench with Zephy-7B model outperforming majority of existing LLMs with similar parameter size. Curry-DPO also achieves the highest adjusted win rates on Vicuna, WizardLM, and UltraFeedback test datasets (90.7%, 87.1%, and 87.9% respectively) in our experiments, with notable gains of upto 7.5% when compared to standard DPO technique.

Reinforcement Learning from Human Feedback (RLHF) aligns language models to human preferences by employing a singular reward model derived from preference data. However, such an approach overlooks the rich diversity of human preferences inherent in data collected from multiple users. In this work, we first derive an impossibility result of alignment with single reward RLHF, thereby highlighting its insufficiency in representing diverse human preferences. To provide an equitable solution to the problem, we learn a mixture of preference distributions via an expectation-maximization algorithm and propose a MaxMin alignment objective for policy learning inspired by the Egalitarian principle in social choice theory to better represent diverse human preferences. We elucidate the connection of our proposed approach to distributionally robust optimization and general utility RL, thereby highlighting the generality and robustness of our proposed solution. We present comprehensive experimental results on small-scale (GPT-2) and large-scale language models (with Tulu2-7B) and show the efficacy of the proposed approach in the presence of diversity among human preferences. Our algorithm achieves an average improvement of more than 16% in win-rates over conventional RLHF algorithms and improves the win-rate (accuracy) for minority groups by over 33% without compromising the performance of majority groups, showcasing the robustness and fairness of our approach. We remark that our findings in this work are not only limited to language models but also extend to reinforcement learning in general.

The alignment of large language models (LLMs) often assumes that using more clean data yields better outcomes, overlooking the match between model capacity and example difficulty. Challenging this, we propose a new principle: Preference data vary in difficulty, and overly difficult examples hinder alignment, by exceeding the model's capacity. Through systematic experimentation, we validate this principle with three key findings: (1) preference examples vary in difficulty, as evidenced by consistent learning orders across alignment runs; (2) overly difficult examples significantly degrade performance across four LLMs and two datasets; and (3) the capacity of a model dictates its threshold for handling difficult examples, underscoring a critical relationship between data selection and model capacity. Building on this principle, we introduce Selective DPO, which filters out overly difficult examples. This simple adjustment improves alignment performance by 9-16% in win rates on the AlpacaEval 2 benchmark compared to the DPO baseline, suppressing a series of DPO variants with different algorithmic adjustments. Together, these results illuminate the importance of aligning data difficulty with model capacity, offering a transformative perspective for improving alignment strategies in LLMs. Code is available at https://github.com/glorgao/SelectiveDPO.

We propose a new method, instruction back-and-forth translation, to construct high-quality synthetic data grounded in world knowledge for aligning large language models (LLMs). Given documents from a web corpus, we generate and curate synthetic instructions using the backtranslation approach proposed by Li et al.(2023a), and rewrite the responses to improve their quality further based on the initial documents. Fine-tuning with the resulting (backtranslated instruction, rewritten response) pairs yields higher win rates on AlpacaEval than using other common instruction datasets such as Humpback, ShareGPT, Open Orca, Alpaca-GPT4 and Self-instruct. We also demonstrate that rewriting the responses with an LLM outperforms direct distillation, and the two generated text distributions exhibit significant distinction in embedding space. Further analysis shows that our backtranslated instructions are of higher quality than other sources of synthetic instructions, while our responses are more diverse and complex than those obtained from distillation. Overall we find that instruction back-and-forth translation combines the best of both worlds -- making use of the information diversity and quantity found on the web, while ensuring the quality of the responses which is necessary for effective alignment.

When prompting a language model (LM), users frequently expect the model to adhere to a set of behavioral principles across diverse tasks, such as producing insightful content while avoiding harmful or biased language. Instilling such principles into a model can be resource-intensive and technically challenging, generally requiring human preference labels or examples. We introduce SAMI, a method for teaching a pretrained LM to follow behavioral principles that does not require any preference labels or demonstrations. SAMI is an iterative algorithm that finetunes a pretrained LM to increase the conditional mutual information between constitutions and self-generated responses given queries from a datasest. On single-turn dialogue and summarization, a SAMI-trained mistral-7b outperforms the initial pretrained model, with win rates between 66% and 77%. Strikingly, it also surpasses an instruction-finetuned baseline (mistral-7b-instruct) with win rates between 55% and 57% on single-turn dialogue. SAMI requires a "principle writer" model; to avoid dependence on stronger models, we further evaluate aligning a strong pretrained model (mixtral-8x7b) using constitutions written by a weak instruction-finetuned model (mistral-7b-instruct). The SAMI-trained mixtral-8x7b outperforms both the initial model and the instruction-finetuned model, achieving a 65% win rate on summarization. Our results indicate that a pretrained LM can learn to follow constitutions without using preference labels, demonstrations, or human oversight.

Long-context dialogue systems suffer from State Inertia, where static constraints prevent models from resolving conflicts between evolving user intents and established historical context. To address this, we propose DZ-TDPO, a non-destructive alignment framework that synergizes conflict-aware dynamic KL constraints with a calibrated temporal attention bias. Experiments on the Multi-Session Chat (MSC) dataset demonstrate that DZ-TDPO achieves state-of-the-art win rates (55.4% on Phi-3.5) while maintaining robust zero-shot generalization. Our scaling analysis reveals a "Capacity-Stability Trade-off": while smaller models incur an "alignment tax" (perplexity surge) to overcome historical inertia, the larger Qwen2.5-7B model achieves 50.8% win rate with negligible perplexity overhead. This confirms that TAI can be alleviated via precise attention regulation rather than destructive weight updates, preserving general capabilities (MMLU) across model scales. Code and data are available: https://github.com/lyj20071013/DZ-TDPO

Language model alignment has become a critical step in training modern generative language models. The goal of alignment is to finetune a reference model such that the win rate of a sample from the aligned model over a sample from the reference model is high, subject to a KL divergence constraint. Today, we are increasingly using inference-time algorithms (e.g., Best-of-N, controlled decoding, tree search) to decode from language models rather than standard sampling. However, the alignment objective does not capture such inference-time decoding procedures. We show that the existing alignment framework is sub-optimal in view of such inference-time methods. We then modify the alignment objective and propose a framework for inference-aware alignment (IAPO). We prove that for any inference-time decoding algorithm, the optimal solution that optimizes the inference-time win rate of the aligned policy against the reference policy is the solution to the typical RLHF problem with a transformation of the reward. This motivates us to provide the KL-regularized calibrate-and-transform RL (CTRL) algorithm to solve this problem, which involves a reward calibration step and a KL-regularized reward maximization step with a transformation of the calibrated reward. We particularize our study to two important inference-time strategies: best-of-N sampling and best-of-N jailbreaking, where N responses are sampled from the model and the one with the highest or lowest reward is selected. We propose specific transformations for these strategies and demonstrate that our framework offers significant improvements over existing state-of-the-art methods for language model alignment. Empirically, we outperform baselines that are designed without taking inference-time decoding into consideration by 8-12% and 4-9% on inference-time win rates over the Anthropic helpfulness and harmlessness dialog benchmark datasets.

Aligning large language models with human preferences is critical for creating reliable and controllable AI systems. A human preference can be visualized as a high-dimensional vector where different directions represent trade-offs between desired attributes (e.g., helpfulness vs. verbosity). Yet, because the training data often reflects dominant, average preferences, LLMs tend to perform well on common requests but fall short in specific, individual needs. This mismatch creates a preference coverage gap. Existing methods often address this through costly retraining, which may not be generalized to the full spectrum of diverse preferences. This brittleness means that when a user's request reflects a nuanced preference deviating from the training data's central tendency, model performance can degrade unpredictably. To address this challenge, we introduce Robust Preference Selection (RPS), a post-hoc, training-free method by leveraging directional neighborhood consensus. Instead of forcing a model to generate a response from a single, highly specific preference, RPS samples multiple responses from a local neighborhood of related preferences to create a superior candidate pool. It then selects the response that best aligns with the user's original intent. We provide a theoretical framework showing our neighborhood generation strategy is provably superior to a strong baseline that also samples multiple candidates. Comprehensive experiments across three distinct alignment paradigms (DPA, DPO, and SFT) demonstrate that RPS consistently improves robustness against this baseline, achieving win rates of up to 69% on challenging preferences from under-represented regions of the space without any model retraining. Our work presents a practical, theoretically-grounded solution for enhancing the reliability of preference-aligned models.

Self-play alignment algorithms have been developed as effective methods for fine-tuning large language models (LLMs), formulating preference optimization as a two-player game. However, the regularization with respect to the reference policy, which is crucial for mitigating over-optimization, has been insufficiently investigated in self-play alignment. In this paper, we show that our regularization method can improve the unregularized self-play significantly. To study the impact of different regularizations in self-play alignment, we propose Regularized Self-Play Policy Optimization (RSPO). This generalized framework regularizes the self-play by simply adding a chosen regularization term into the loss while maintaining provable last-iterate convergence to the Nash Equilibrium of the corresponding regularized game. Surprisingly, empirical evaluations using the Mistral-7B-Instruct base model reveal that forward KL divergence regularization reduces response length in RSPO, whereas reverse KL divergence markedly improves raw win rates. RSPO with a linear combination of forward and reverse KL divergence regularization substantially increases the length-controlled win rate in AlpacaEval-2, elevating the unregularized self-play alignment method (SPPO) from 28.53% to 35.44%. Finally, we show that RSPO also improves the response diversity.

Limited access to mental healthcare, extended wait times, and increasing capabilities of Large Language Models (LLMs) has led individuals to turn to LLMs for fulfilling their mental health needs. However, examining the multi-turn mental health conversation capabilities of LLMs remains under-explored. Existing evaluation frameworks typically focus on diagnostic accuracy and win-rates and often overlook alignment with patient-specific goals, values, and personalities required for meaningful conversations. To address this, we introduce MedAgent, a novel framework for synthetically generating realistic, multi-turn mental health sensemaking conversations and use it to create the Mental Health Sensemaking Dialogue (MHSD) dataset, comprising over 2,200 patient-LLM conversations. Additionally, we present MultiSenseEval, a holistic framework to evaluate the multi-turn conversation abilities of LLMs in healthcare settings using human-centric criteria. Our findings reveal that frontier reasoning models yield below-par performance for patient-centric communication and struggle at advanced diagnostic capabilities with average score of 31%. Additionally, we observed variation in model performance based on patient's persona and performance drop with increasing turns in the conversation. Our work provides a comprehensive synthetic data generation framework, a dataset and evaluation framework for assessing LLMs in multi-turn mental health conversations.

Direct Preference Optimisation (DPO) has emerged as a powerful method for aligning Large Language Models (LLMs) with human preferences, offering a stable and efficient alternative to approaches that use Reinforcement learning via Human Feedback. In this work, we investigate the performance of DPO using open-source preference datasets. One of the major drawbacks of DPO is that it doesn't induce granular scoring and treats all the segments of the responses with equal propensity. However, this is not practically true for human preferences since even "good" responses have segments that may not be preferred by the annotator. To resolve this, a 2-dimensional scoring for DPO alignment called 2D-DPO was proposed. We explore the 2D-DPO alignment paradigm and the advantages it provides over the standard DPO by comparing their win rates. It is observed that these methods, even though effective, are not robust to label/score noise. To counter this, we propose an approach of incorporating segment-level score noise robustness to the 2D-DPO algorithm. Along with theoretical backing, we also provide empirical verification in favour of the algorithm and introduce other noise models that can be present.

Reinforcement Learning from Human/AI Feedback (RLHF/RLAIF) has been extensively utilized for preference alignment of text-to-image models. Existing methods face certain limitations in terms of both data and algorithm. For training data, most approaches rely on manual annotated preference data, either by directly fine-tuning the generators or by training reward models to provide training signals. However, the high annotation cost makes them difficult to scale up, the reward model consumes extra computation and cannot guarantee accuracy. From an algorithmic perspective, most methods neglect the value of text and only take the image feedback as a comparative signal, which is inefficient and sparse. To alleviate these drawbacks, we propose the InstructEngine framework. Regarding annotation cost, we first construct a taxonomy for text-to-image generation, then develop an automated data construction pipeline based on it. Leveraging advanced large multimodal models and human-defined rules, we generate 25K text-image preference pairs. Finally, we introduce cross-validation alignment method, which refines data efficiency by organizing semantically analogous samples into mutually comparable pairs. Evaluations on DrawBench demonstrate that InstructEngine improves SD v1.5 and SDXL's performance by 10.53% and 5.30%, outperforming state-of-the-art baselines, with ablation study confirming the benefits of InstructEngine's all components. A win rate of over 50% in human reviews also proves that InstructEngine better aligns with human preferences.

Inference-time alignment enhances the performance of large language models without requiring additional training or fine-tuning but presents challenges due to balancing computational efficiency with high-quality output. Best-of-N (BoN) sampling, as a simple yet powerful approach, generates multiple responses and selects the best one, achieving improved performance but with a high computational cost. We propose TreeBoN, a novel framework that integrates a speculative tree-search strategy into Best-of-N (BoN) Sampling. TreeBoN maintains a set of parent nodes, iteratively branching and pruning low-quality responses, thereby reducing computational overhead while maintaining high output quality. Our approach also leverages token-level rewards from Direct Preference Optimization (DPO) to guide tree expansion and prune low-quality paths. We evaluate TreeBoN using AlpacaFarm, UltraFeedback, GSM8K, HH-RLHF, and TutorEval datasets, demonstrating consistent improvements. Specifically, TreeBoN achieves a 65% win rate at maximum lengths of 192 and 384 tokens, outperforming standard BoN with the same computational cost. Furthermore, TreeBoN achieves around a 60% win rate across longer responses, showcasing its scalability and alignment efficacy.

Building helpful and harmless large language models (LLMs) requires effective model alignment approach based on human instructions and feedback, which necessitates high-quality human-labeled data. Constructing such datasets is often expensive and hard to scale, and may face potential limitations on diversity and generalization. To address these challenges, we introduce Mixture of Agents Alignment (MoAA), that leverages the collective strengths of various language models to provide high-quality data for model alignment. By employing MoAA, we enhance both supervised fine-tuning and preference optimization, leading to improved performance compared to using a single model alone to generate alignment data (e.g. using GPT-4o alone). Evaluation results show that our approach can improve win rate of LLaMA-3.1-8B-Instruct from 19.5 to 48.3 on Arena-Hard and from 22.33 to 57.23 on AlpacaEval2, highlighting a promising direction for model alignment through this new scalable and diverse synthetic data recipe. Furthermore, we demonstrate that MoAA enables a self-improvement pipeline, where models finetuned on MoA-generated data surpass their own initial capabilities, providing evidence that our approach can push the frontier of open-source LLMs without reliance on stronger external supervision. Data and code will be released.

Current reinforcement learning (RL) frameworks for large language models (LLM) post-training typically assume a fixed prompt distribution, which is sub-optimal and bottlenecks scalability. Prior works have explored prompt evolving, but are often limited to the supervised fine-tuning stage, and prompts are sampled and evolved uniformly without signals. This empirical work presents a paradigm shift: Evolving Alignment via Asymmetric Self-Play (eva), that casts post-training as an infinite game with regret-based signals for 2 players: (i) a creator, who strategically samples and creates new informative prompts and (ii) a solver, who learns to produce preferred responses. eva is the first method that allows language models to adaptively create training prompts in both offline and online RL post-training. The design is simple, easy-to-use yet remarkably effective: eva sets a new SOTA on challenging benchmarks, without any extra human prompts, e.g. it boosts the win-rate of gemma-2-9b-it on Arena-Hard by 51.6% -> 60.1% for DPO and 52.6% -> 62.4% for RLOO, surpassing claude-3-opus and catching up to gemini-1.5-pro, both of which are orders of magnitude larger. Extensive experiments show eva can create effective RL curricula and is robust across ablations. We believe adaptively evolving prompts are key to designing the next-generation RL post-training scheme.

Aligning large language models (LLMs) to human preferences is a crucial step in building helpful and safe AI tools, which usually involve training on supervised datasets. Popular algorithms such as Direct Preference Optimization rely on pairs of AI-generated responses ranked according to human feedback. The labeling process is the most labor-intensive and costly part of the alignment pipeline, and improving its efficiency would have a meaningful impact on AI development. We propose a strategy for sampling a high-quality training dataset that focuses on acquiring the most informative response pairs for labeling out of a set of AI-generated responses. Experimental results on synthetic HH-RLHF benchmarks indicate that choosing dissimilar response pairs enhances the direct alignment of LLMs while reducing inherited labeling errors. We also applied our method to the real-world dataset SHP2, selecting optimal pairs from multiple responses. The model aligned on dissimilar response pairs obtained the best win rate on the dialogue task. Our findings suggest that focusing on less similar pairs can improve the efficiency of LLM alignment, saving up to 65% of annotators' work.

Modern language models often rely on Reinforcement Learning from Human Feedback (RLHF) to encourage safe behaviors. However, they remain vulnerable to adversarial attacks due to three key limitations: (1) the inefficiency and high cost of human annotation, (2) the vast diversity of potential adversarial attacks, and (3) the risk of feedback bias and reward hacking. To address these challenges, we introduce Adversarial Preference Learning (APL), an iterative adversarial training method incorporating three key innovations. First, a direct harmfulness metric based on the model's intrinsic preference probabilities, eliminating reliance on external assessment. Second, a conditional generative attacker that synthesizes input-specific adversarial variations. Third, an iterative framework with automated closed-loop feedback, enabling continuous adaptation through vulnerability discovery and mitigation. Experiments on Mistral-7B-Instruct-v0.3 demonstrate that APL significantly enhances robustness, achieving 83.33% harmlessness win rate over the base model (evaluated by GPT-4o), reducing harmful outputs from 5.88% to 0.43% (measured by LLaMA-Guard), and lowering attack success rate by up to 65% according to HarmBench. Notably, APL maintains competitive utility, with an MT-Bench score of 6.59 (comparable to the baseline 6.78) and an LC-WinRate of 46.52% against the base model.

This paper concerns the problem of aligning samples from large language models to human preferences using best-of-n sampling, where we draw n samples, rank them, and return the best one. We consider two fundamental problems. First: what is the relationship between best-of-n and approaches to alignment that train LLMs to output samples with a high expected reward (e.g., RLHF or DPO)? To answer this, we embed both the best-of-n distribution and the sampling distributions learned by alignment procedures in a common class of tiltings of the base LLM distribution. We then show that, within this class, best-of-n is essentially optimal in terms of the trade-off between win-rate against the base model vs KL distance from the base model. That is, best-of-n is the best choice of alignment distribution if the goal is to maximize win rate. However, best-of-n requires drawing n samples for each inference, a substantial cost. To avoid this, the second problem we consider is how to fine-tune a LLM to mimic the best-of-n sampling distribution. We derive BoNBoN Alignment to achieve this by exploiting the special structure of the best-of-n distribution. Experiments show that BoNBoN alignment yields substantial improvements in producing a model that is preferred to the base policy while minimally affecting off-target aspects.

Text-to-image diffusion models have revolutionized visual content generation, but current safety mechanisms apply uniform standards that often fail to account for individual user preferences. These models overlook the diverse safety boundaries shaped by factors like age, mental health, and personal beliefs. To address this, we propose Personalized Safety Alignment (PSA), a framework that allows user-specific control over safety behaviors in generative models. PSA integrates personalized user profiles into the diffusion process, adjusting the model's behavior to match individual safety preferences while preserving image quality. We introduce a new dataset, Sage, which captures user-specific safety preferences and incorporates these profiles through a cross-attention mechanism. Experiments show that PSA outperforms existing methods in harmful content suppression and aligns generated content better with user constraints, achieving higher Win Rate and Pass Rate scores. Our code, data, and models are publicly available at https://torpedo2648.github.io/PSAlign/.

Standard human preference-based alignment methods, such as Reinforcement Learning from Human Feedback (RLHF), are a cornerstone technology for aligning Large Language Models (LLMs) with human values. However, these methods are all underpinned by a critical, yet flawed assumption: human preferences are homogeneous (representing a single, unified preference) and the collected data is noiseless (free from error). In reality, neither is true since human preference is pluralistic and annotators can make mistakes. This creates a discrepancy between the recorded data and the ground-truth preferences, which can misguide the model and degrade its performance. To address this challenge, we introduce Latent Collective Preference Optimization (LCPO). LCPO leverages an Expectation-Maximization (EM) algorithm to learn the latent collective consensus from noisy data. It operates by inferring the correctness of each preference label and using this probability as an adaptive weight to re-calibrate each data point's contribution to the training loss, thereby mitigating noise. We generalize this approach by establishing a theoretical link between arbitrary preference losses and their corresponding probabilistic models, elevating LCPO from a specific algorithm to a general framework for robust preference alignment. Theoretically, we prove that under the condition of a perfectly calibrated model, LCPO is guaranteed to converge to the true noise level of the dataset. Our experiments demonstrate LCPO's effectiveness as a general framework, consistently enhancing four state-of-the-art alignment algorithms (DPO, IPO, SimPO, and CPO). When applied to Mistral and Llama 3 models, the LCPO-enhanced methods achieve substantial win rate gains on AlpacaEval 2 and Arena-Hard, with improvements of up to 7.0% on both benchmarks.

Learning from human feedback is a prominent technique to align the output of large language models (LLMs) with human expectations. Reinforcement learning from human feedback (RLHF) leverages human preference signals that are in the form of ranking of response pairs to perform this alignment. However, human preference on LLM outputs can come in much richer forms including natural language, which may provide detailed feedback on strengths and weaknesses of a given response. In this work we investigate data efficiency of modeling human feedback that is in natural language. Specifically, we fine-tune an open-source LLM, e.g., Falcon-40B-Instruct, on a relatively small amount (1000 records or even less) of human feedback in natural language in the form of critiques and revisions of responses. We show that this model is able to improve the quality of responses from even some of the strongest LLMs such as ChatGPT, BARD, and Vicuna, through critique and revision of those responses. For instance, through one iteration of revision of ChatGPT responses, the revised responses have 56.6% win rate over the original ones, and this win rate can be further improved to 65.9% after applying the revision for five iterations.

In this note, we explore inference-time alignment through in-context learning. We consider a vanilla pretrained language model Llama-2 before any fine-tuning and retrieve an average of 9 demonstration alignment examples when the model is prompted to follow chat-style instructions. Compared to direct prompting, the in-context alignment without changing model weights leads to a 7x increase in win-rate w.r.t. the text-davinci-003 model from OpenAI, making the vanilla language model comparable to strong baselines with alignment fine-tuning.

Large Language Models (LLMs) are expected to produce safe, helpful, and honest content during interaction with human users, but they frequently fail to align with such values when given flawed instructions, e.g., missing context, ambiguous directives, or inappropriate tone, leaving substantial room for improvement along multiple dimensions. A cost-effective yet high-impact way is to pre-align instructions before the model begins decoding. Existing approaches either rely on prohibitive test-time search costs or end-to-end model rewrite, which is powered by a customized training corpus with unclear objectives. In this work, we demonstrate that the goal of efficient and effective preference alignment can be achieved by P-Aligner, a lightweight module generating instructions that preserve the original intents while being expressed in a more human-preferred form. P-Aligner is trained on UltraPrompt, a new dataset synthesized via a proposed principle-guided pipeline using Monte-Carlo Tree Search, which systematically explores the space of candidate instructions that are closely tied to human preference. Experiments across different methods show that P-Aligner generally outperforms strong baselines across various models and benchmarks, including average win-rate gains of 28.35% and 8.69% on GPT-4-turbo and Gemma-2-SimPO, respectively. Further analyses validate its effectiveness and efficiency through multiple perspectives, including data quality, search strategies, iterative deployment, and time overhead.

Visual layouts are essential in graphic design fields such as advertising, posters, and web interfaces. The application of generative models for content-aware layout generation has recently gained traction. However, these models fail to understand the contextual aesthetic requirements of layout design and do not align with human-like preferences, primarily treating it as a prediction task without considering the final rendered output. To overcome these problems, we offer Aesthetic-Aware Preference Alignment(AAPA), a novel technique to train a Multi-modal Large Language Model (MLLM) for layout prediction that uses MLLM's aesthetic preferences for Direct Preference Optimization over graphic layouts. We propose a data filtering protocol utilizing our layout-quality heuristics for AAPA to ensure training happens on high-quality layouts. Additionally, we introduce a novel evaluation metric that uses another MLLM to compute the win rate of the generated layout against the ground-truth layout based on aesthetics criteria. We also demonstrate the applicability of AAPA for MLLMs of varying scales (1B to 8B parameters) and LLM families (Qwen, Phi, InternLM). By conducting thorough qualitative and quantitative analyses, we verify the efficacy of our approach on two challenging benchmarks - Crello and Webui, showcasing 17%, and 16 improvement over current State-of-The-Art methods, thereby highlighting the potential of MLLMs in aesthetic-aware layout generation.

Large Language Models (LLMs) are rapidly surpassing human knowledge in many domains. While improving these models traditionally relies on costly human data, recent self-rewarding mechanisms (Yuan et al., 2024) have shown that LLMs can improve by judging their own responses instead of relying on human labelers. However, existing methods have primarily focused on improving model responses rather than judgment capabilities, resulting in rapid saturation during iterative training. To address this issue, we introduce a novel Meta-Rewarding step to the self-improvement process, where the model judges its own judgements and uses that feedback to refine its judgment skills. Surprisingly, this unsupervised approach improves the model's ability to judge {\em and} follow instructions, as demonstrated by a win rate improvement of Llama-3-8B-Instruct from 22.9% to 39.4% on AlpacaEval 2, and 20.6% to 29.1% on Arena-Hard. These results strongly suggest the potential for self-improving models without human supervision.

The OpenAI o1-series models have demonstrated that leveraging long-form Chain of Thought (CoT) can substantially enhance performance. However, the recursive thinking capabilities of Large Language Models (LLMs) remain limited, particularly in the absence of expert-curated data for distillation. In this paper, we propose AvR: Alignment via Refinement, a novel method aimed at unlocking the potential of LLMs for recursive reasoning through long-form CoT. AvR introduces a refinement process that integrates criticism and improvement actions, guided by differentiable learning techniques to optimize refinement-aware rewards. As a result, the synthesized multi-round data can be organized as a long refinement thought, further enabling test-time scaling. Experimental results show that AvR significantly outperforms conventional preference optimization methods. Notably, with only 3k synthetic samples, our method boosts the performance of the LLaMA-3-8B-Instruct model by over 20\% in win rate on AlpacaEval 2.0. Our code is available at Github (https://github.com/Banner-Z/AvR.git).

Traditional reinforcement learning from human feedback (RLHF) approaches relying on parametric models like the Bradley-Terry model fall short in capturing the intransitivity and irrationality in human preferences. Recent advancements suggest that directly working with preference probabilities can yield a more accurate reflection of human preferences, enabling more flexible and accurate language model alignment. In this paper, we propose a self-play-based method for language model alignment, which treats the problem as a constant-sum two-player game aimed at identifying the Nash equilibrium policy. Our approach, dubbed Self-Play Preference Optimization (SPPO), approximates the Nash equilibrium through iterative policy updates and enjoys theoretical convergence guarantee. Our method can effectively increase the log-likelihood of the chosen response and decrease that of the rejected response, which cannot be trivially achieved by symmetric pairwise loss such as Direct Preference Optimization (DPO) and Identity Preference Optimization (IPO). In our experiments, using only 60k prompts (without responses) from the UltraFeedback dataset and without any prompt augmentation, by leveraging a pre-trained preference model PairRM with only 0.4B parameters, SPPO can obtain a model from fine-tuning Mistral-7B-Instruct-v0.2 that achieves the state-of-the-art length-controlled win-rate of 28.53% against GPT-4-Turbo on AlpacaEval 2.0. It also outperforms the (iterative) DPO and IPO on MT-Bench and the Open LLM Leaderboard. Notably, the strong performance of SPPO is achieved without additional external supervision (e.g., responses, preferences, etc.) from GPT-4 or other stronger language models.

Human alignment in large language models (LLMs) is an active area of research. A recent groundbreaking work, direct preference optimization (DPO), has greatly simplified the process from past work in reinforcement learning from human feedback (RLHF) by bypassing the reward learning stage in RLHF. DPO, after training, provides an implicit reward model. In this work, we make a novel observation that this implicit reward model can by itself be used in a bootstrapping fashion to further align the LLM. Our approach is to use the rewards from a current LLM model to construct a preference dataset, which is then used in subsequent DPO rounds. We incorporate refinements that debias the length of the responses and improve the quality of the preference dataset to further improve our approach. Our approach, named self-alignment with DPO ImpliCit rEwards (DICE), shows great improvements in alignment and achieves superior performance than Gemini Pro on AlpacaEval 2, reaching 27.55% length-controlled win rate against GPT-4 Turbo, but with only 8B parameters and no external feedback. Our code is available at https://github.com/sail-sg/dice.

Language models must be adapted to understand and follow user instructions. Reinforcement learning is widely used to facilitate this -- typically using fixed criteria such as "helpfulness" and "harmfulness". In our work, we instead propose using flexible, instruction-specific criteria as a means of broadening the impact that reinforcement learning can have in eliciting instruction following. We propose "Reinforcement Learning from Checklist Feedback" (RLCF). From instructions, we extract checklists and evaluate how well responses satisfy each item - using both AI judges and specialized verifier programs - then combine these scores to compute rewards for RL. We compare RLCF with other alignment methods applied to a strong instruction following model (Qwen2.5-7B-Instruct) on five widely-studied benchmarks -- RLCF is the only method to improve performance on every benchmark, including a 4-point boost in hard satisfaction rate on FollowBench, a 6-point increase on InFoBench, and a 3-point rise in win rate on Arena-Hard. These results establish checklist feedback as a key tool for improving language models' support of queries that express a multitude of needs.

In this study, we explore Transformer-based diffusion models for image and video generation. Despite the dominance of Transformer architectures in various fields due to their flexibility and scalability, the visual generative domain primarily utilizes CNN-based U-Net architectures, particularly in diffusion-based models. We introduce GenTron, a family of Generative models employing Transformer-based diffusion, to address this gap. Our initial step was to adapt Diffusion Transformers (DiTs) from class to text conditioning, a process involving thorough empirical exploration of the conditioning mechanism. We then scale GenTron from approximately 900M to over 3B parameters, observing significant improvements in visual quality. Furthermore, we extend GenTron to text-to-video generation, incorporating novel motion-free guidance to enhance video quality. In human evaluations against SDXL, GenTron achieves a 51.1% win rate in visual quality (with a 19.8% draw rate), and a 42.3% win rate in text alignment (with a 42.9% draw rate). GenTron also excels in the T2I-CompBench, underscoring its strengths in compositional generation. We believe this work will provide meaningful insights and serve as a valuable reference for future research.

City-scale 3D generation is of great importance for the development of embodied intelligence and world models. Existing methods, however, face significant challenges regarding quality, fidelity, and scalability in 3D world generation. Thus, we propose RAISECity, a Reality-Aligned Intelligent Synthesis Engine that creates detailed, City-scale 3D worlds. We introduce an agentic framework that leverages diverse multimodal foundation tools to acquire real-world knowledge, maintain robust intermediate representations, and construct complex 3D scenes. This agentic design, featuring dynamic data processing, iterative self-reflection and refinement, and the invocation of advanced multimodal tools, minimizes cumulative errors and enhances overall performance. Extensive quantitative experiments and qualitative analyses validate the superior performance of RAISECity in real-world alignment, shape precision, texture fidelity, and aesthetics level, achieving over a 90% win-rate against existing baselines for overall perceptual quality. This combination of 3D quality, reality alignment, scalability, and seamless compatibility with computer graphics pipelines makes RAISECity a promising foundation for applications in immersive media, embodied intelligence, and world models.

Despite rapid advances in text-to-video synthesis, generated video quality remains critically dependent on precise user prompts. Existing test-time optimization methods, successful in other domains, struggle with the multi-faceted nature of video. In this work, we introduce VISTA (Video Iterative Self-improvemenT Agent), a novel multi-agent system that autonomously improves video generation through refining prompts in an iterative loop. VISTA first decomposes a user idea into a structured temporal plan. After generation, the best video is identified through a robust pairwise tournament. This winning video is then critiqued by a trio of specialized agents focusing on visual, audio, and contextual fidelity. Finally, a reasoning agent synthesizes this feedback to introspectively rewrite and enhance the prompt for the next generation cycle. Experiments on single- and multi-scene video generation scenarios show that while prior methods yield inconsistent gains, VISTA consistently improves video quality and alignment with user intent, achieving up to 60% pairwise win rate against state-of-the-art baselines. Human evaluators concur, preferring VISTA outputs in 66.4% of comparisons.

The alignment of large language models (LLMs) with human preferences remains a key challenge. While post-training techniques like Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO) have achieved notable success, they often introduce computational inefficiencies and training instability. In this paper, we propose Feature-level constrained Preference Optimization (FPO), a novel method designed to simplify the alignment process while ensuring stability. FPO leverages pre-trained Sparse Autoencoders (SAEs) and introduces feature-level constraints, allowing for efficient, sparsity-enforced alignment. Our approach enjoys efficiency by using sparse features activated in a well-trained sparse autoencoder and the quality of sequential KL divergence by using the feature-level offline reference. Experimental results on benchmark datasets demonstrate that FPO achieves a 5.08% absolute improvement in win rate with much lower computational cost compared to state-of-the-art baselines, making it a promising solution for efficient and controllable LLM alignments.

Direct Preference Optimization (DPO) has become a popular approach for aligning language models using pairwise preferences. However, in practical post-training pipelines, on-policy generation typically yields multiple candidate responses per prompt, which are scored by a reward model to guide learning. In this setting, we propose Multi-Preference Optimization (MPO), a generalization of DPO that optimizes over entire sets of responses by extending the Bradley-Terry model to groupwise comparisons between chosen and rejected sets. To further enhance learning, MPO employs deviation-based weighting, which emphasizes outlier responses that deviate most from the mean reward, effectively inducing a self-paced curriculum. We theoretically prove that MPO reduces alignment bias at a rate of Oleft(1{n}right) with respect to the number of responses per query. Empirically, MPO achieves state-of-the-art performance on the UltraFeedback benchmark and yields up to sim 17.5% improvement over the state-of-the-art baseline in length-controlled win rate on AlpacaEval2, establishing a new baseline for preference-based alignment

Aligning large language models (LLMs) with human preferences through reinforcement learning (RLHF) can lead to reward hacking, where LLMs exploit failures in the reward model (RM) to achieve seemingly high rewards without meeting the underlying objectives. We identify two primary challenges when designing RMs to mitigate reward hacking: distribution shifts during the RL process and inconsistencies in human preferences. As a solution, we propose Weight Averaged Reward Models (WARM), first fine-tuning multiple RMs, then averaging them in the weight space. This strategy follows the observation that fine-tuned weights remain linearly mode connected when sharing the same pre-training. By averaging weights, WARM improves efficiency compared to the traditional ensembling of predictions, while improving reliability under distribution shifts and robustness to preference inconsistencies. Our experiments on summarization tasks, using best-of-N and RL methods, shows that WARM improves the overall quality and alignment of LLM predictions; for example, a policy RL fine-tuned with WARM has a 79.4% win rate against a policy RL fine-tuned with a single RM.

We propose SPARTA ALIGNMENT, an algorithm to collectively align multiple LLMs through competition and combat. To complement a single model's lack of diversity in generation and biases in evaluation, multiple LLMs form a "sparta tribe" to compete against each other in fulfilling instructions while serving as judges for the competition of others. For each iteration, one instruction and two models are selected for a duel, the other models evaluate the two responses, and their evaluation scores are aggregated through a adapted elo-ranking based reputation system, where winners/losers of combat gain/lose weight in evaluating others. The peer-evaluated combat results then become preference pairs where the winning response is preferred over the losing one, and all models learn from these preferences at the end of each iteration. SPARTA ALIGNMENT enables the self-evolution of multiple LLMs in an iterative and collective competition process. Extensive experiments demonstrate that SPARTA ALIGNMENT outperforms initial models and 4 self-alignment baselines across 10 out of 12 tasks and datasets with 7.0% average improvement. Further analysis reveals that SPARTA ALIGNMENT generalizes more effectively to unseen tasks and leverages the expertise diversity of participating models to produce more logical, direct and informative outputs.

Mainstream approaches to aligning large language models (LLMs) heavily rely on human preference data, particularly when models require periodic updates. The standard process for iterative alignment of LLMs involves collecting new human feedback for each update. However, the data collection process is costly and challenging to scale. To address this issue, we introduce the "TS-Align" framework, which fine-tunes a policy model using pairwise feedback data automatically mined from its outputs. This automatic mining process is efficiently accomplished through the collaboration between a large-scale teacher model and a small-scale student model. The policy fine-tuning process can be iteratively repeated using on-policy generations within our proposed teacher-student collaborative framework. Through extensive experiments, we demonstrate that our final aligned policy outperforms the base policy model with an average win rate of 69.7% across seven conversational or instruction-following datasets. Furthermore, we show that the ranking capability of the teacher is effectively distilled into the student through our pipeline, resulting in a small-scale yet effective reward model for policy model alignment.

Alignment has quickly become a default ingredient in LLM development, with techniques such as reinforcement learning from human feedback making models act safely, follow instructions, and perform ever-better on complex tasks. While these techniques are certainly useful, we propose that they should not be universally applied and demonstrate a range of tasks on which base language models consistently outperform their popular aligned forms. Particularly, we study tasks that require unpredictable outputs, such as random number generation, mixed strategy games (rock-paper-scissors and hide-and-seek), and creative writing. In each case, aligned models tend towards narrow behaviors that result in distinct disadvantages, for instance, preferring to generate "7" over other uniformly random numbers, becoming almost fully predictable in some game states, or prioritizing pleasant writing over creative originality. Across models tested, better performance on common benchmarks tends to correlate with worse performance on our tasks, suggesting an effective trade-off in the required capabilities.

Although humans inherently have diverse values, current large language model (LLM) alignment methods often assume that aligning LLMs with the general public's preferences is optimal. A major challenge in adopting a more individualized approach to LLM alignment is its lack of scalability, as it involves repeatedly acquiring preference data and training new reward models and LLMs for each individual's preferences. To address these challenges, we propose a new paradigm where users specify what they value most within the system message, steering the LLM's generation behavior to better align with the user's intentions. However, a naive application of such an approach is non-trivial since LLMs are typically trained on a uniform system message (e.g., "You are a helpful assistant") which limits their ability to generalize to diverse, unseen system messages. To improve this generalization, we create the Multifaceted Collection, a preference dataset with 192k combinations of values beyond generic helpfulness and harmlessness, spanning 65k user instructions. Using this dataset, we train a 7B LLM called Janus and test it on 921 prompts from 5 benchmarks (AlpacaEval 2.0, FLASK, Koala, MT-Bench, and Self-Instruct) by adding various unseen system messages that reflect user preferences. Janus achieves tie+win rate of 75.2%, 72.4%, and 66.4% against Mistral 7B Instruct v0.2, GPT-3.5 Turbo, and GPT-4, respectively. Unexpectedly, on three benchmarks focused on response helpfulness (AlpacaEval 2.0, MT-Bench, Arena Hard Auto v0.1), Janus also outperforms LLaMA 3 8B Instruct by a +4.0%, +0.1%, +3.0% margin, underscoring that training with a vast array of system messages could also enhance alignment to the general public's preference as well. Our code, dataset, benchmark, and models are available at https://github.com/kaistAI/Janus.

Post-training alignment optimizes language models to match human preference signals, but this objective is not equivalent to modeling observed human behavior. We compare 120 base-aligned model pairs on more than 10,000 real human decisions in multi-round strategic games - bargaining, persuasion, negotiation, and repeated matrix games. In these settings, base models outperform their aligned counterparts in predicting human choices by nearly 10:1, robustly across model families, prompt formulations, and game configurations. This pattern reverses, however, in settings where human behavior is more likely to follow normative predictions: aligned models dominate on one-shot textbook games across all 12 types tested and on non-strategic lottery choices - and even within the multi-round games themselves, at round one, before interaction history develops. This boundary-condition pattern suggests that alignment induces a normative bias: it improves prediction when human behavior is relatively well captured by normative solutions, but hurts prediction in multi-round strategic settings, where behavior is shaped by descriptive dynamics such as reciprocity, retaliation, and history-dependent adaptation. These results reveal a fundamental trade-off between optimizing models for human use and using them as proxies for human behavior.

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.

Existing work on value alignment typically characterizes value relations statically, ignoring how interventions - such as prompting, fine-tuning, or preference optimization - reshape the broader value system. We introduce the Value Alignment Tax (VAT), a framework that measures how alignment-induced changes propagate across interconnected values relative to achieved on-target gain. VAT captures the dynamics of value expression under alignment pressure. Using a controlled scenario-action dataset grounded in Schwartz value theory, we collect paired pre-post normative judgments and analyze alignment effects across models, values, and alignment strategies. Our results show that alignment often produces uneven, structured co-movement among values. These effects are invisible under conventional target-only evaluation, revealing systemic, process-level alignment risks and offering new insights into the dynamics of value alignment in LLMs.

Direct Alignment Algorithms (DAAs) simplify language model alignment by replacing reinforcement learning (RL) and reward modeling (RM) in Reinforcement Learning from Human Feedback (RLHF) with direct policy optimization. DAAs can be classified by their ranking losses (pairwise vs. pointwise), by the rewards used in those losses (e.g., likelihood ratios of policy and reference policy, or odds ratios), or by whether a Supervised Fine-Tuning (SFT) phase is required (two-stage vs. one-stage). We first show that one-stage methods underperform two-stage methods. To address this, we incorporate an explicit SFT phase and introduce the beta parameter, controlling the strength of preference optimization, into single-stage ORPO and ASFT. These modifications improve their performance in Alpaca Eval 2 by +3.46 (ORPO) and +8.27 (ASFT), matching two-stage methods like DPO. Further analysis reveals that the key factor is whether the approach uses pairwise or pointwise objectives, rather than the specific implicit reward or loss function. These results highlight the importance of careful evaluation to avoid premature claims of performance gains or overall superiority in alignment algorithms.

Frontier AI systems, including large language models (LLMs), hold increasing influence over the epistemology of human users. Such influence can reinforce prevailing societal values, potentially contributing to the lock-in of misguided moral beliefs and, consequently, the perpetuation of problematic moral practices on a broad scale. We introduce progress alignment as a technical solution to mitigate this imminent risk. Progress alignment algorithms learn to emulate the mechanics of human moral progress, thereby addressing the susceptibility of existing alignment methods to contemporary moral blindspots. To empower research in progress alignment, we introduce ProgressGym, an experimental framework allowing the learning of moral progress mechanics from history, in order to facilitate future progress in real-world moral decisions. Leveraging 9 centuries of historical text and 18 historical LLMs, ProgressGym enables codification of real-world progress alignment challenges into concrete benchmarks. Specifically, we introduce three core challenges: tracking evolving values (PG-Follow), preemptively anticipating moral progress (PG-Predict), and regulating the feedback loop between human and AI value shifts (PG-Coevolve). Alignment methods without a temporal dimension are inapplicable to these tasks. In response, we present lifelong and extrapolative algorithms as baseline methods of progress alignment, and build an open leaderboard soliciting novel algorithms and challenges. The framework and the leaderboard are available at https://github.com/PKU-Alignment/ProgressGym and https://huggingface.co/spaces/PKU-Alignment/ProgressGym-LeaderBoard respectively.

AI alignment aims to make AI systems behave in line with human intentions and values. As AI systems grow more capable, so do risks from misalignment. To provide a comprehensive and up-to-date overview of the alignment field, in this survey, we delve into the core concepts, methodology, and practice of alignment. First, we identify four principles as the key objectives of AI alignment: Robustness, Interpretability, Controllability, and Ethicality (RICE). Guided by these four principles, we outline the landscape of current alignment research and decompose them into two key components: forward alignment and backward alignment. The former aims to make AI systems aligned via alignment training, while the latter aims to gain evidence about the systems' alignment and govern them appropriately to avoid exacerbating misalignment risks. On forward alignment, we discuss techniques for learning from feedback and learning under distribution shift. On backward alignment, we discuss assurance techniques and governance practices. We also release and continually update the website (www.alignmentsurvey.com) which features tutorials, collections of papers, blog posts, and other resources.

As large language models (LLMs) are increasingly integrated into social decision-making, understanding their political positioning and alignment behavior is critical for safety and fairness. This study presents a sociotechnical audit of 26 prominent LLMs, triangulating their positions across three psychometric inventories (Political Compass, SapplyValues, 8 Values) and evaluating their performance on a large-scale news labeling task (N approx 27{,}000). Our results reveal a strong clustering of models in the Libertarian-Left region of the ideological space, encompassing 96.3% of the cohort. Alignment signals appear to be consistent architectural traits rather than stochastic noise (η^2 > 0.90); however, we identify substantial discrepancies in measurement validity. In particular, the Political Compass exhibits a strong negative correlation with cultural progressivism (r=-0.64) when compared against multi-axial instruments, suggesting a conflation of social conservatism with authoritarianism in this context. We further observe a significant divergence between open-weights and closed-source models, with the latter displaying markedly higher cultural progressivism scores (p<10^{-25}). In downstream media analysis, models exhibit a systematic "center-shift," frequently categorizing neutral articles as left-leaning, alongside an asymmetric detection capability in which "Far Left" content is identified with greater accuracy (19.2%) than "Far Right" content (2.0%). These findings suggest that single-axis evaluations are insufficient and that multidimensional auditing frameworks are necessary to characterize alignment behavior in deployed LLMs. Our code and data will be made public.

Collecting high-quality preference datasets for reinforcement learning from human feedback (RLHF) is resource-intensive and challenging. As a result, researchers often train reward models on extensive offline datasets which aggregate diverse generation sources and scoring/alignment policies. We hypothesize that this aggregation has an averaging effect on reward model scores, which limits signal and impairs the alignment process. Inspired by the field of inverse RL, we define the 'inverse alignment problem' in language model training, where our objective is to optimize the critic's reward for a fixed actor and a fixed offline preference dataset. We hypothesize that solving the inverse alignment problem will improve reward model quality by providing clearer feedback on the policy's current behavior. To that end, we investigate whether repeatedly fine-tuning a reward model on subsets of the offline preference dataset aligned with a periodically frozen policy during RLHF improves upon vanilla RLHF. Our empirical results demonstrate that this approach facilitates superior alignment and faster convergence compared to using an unaligned or out-of-distribution reward model relative to the LLM policy.

Reinforcement learning from human feedback (RLHF) has emerged as the standard paradigm for aligning large language models (LLMs) with human preferences. However, reward-based methods built on the Bradley-Terry assumption struggle to capture the non-transitive and heterogeneous nature of real-world preferences. To address this, recent studies have reframed alignment as a two-player Nash game, giving rise to Nash learning from human feedback (NLHF). While this perspective has inspired algorithms such as INPO, ONPO, and EGPO with strong theoretical and empirical guarantees, they remain fundamentally restricted to two-player interactions, creating a single-opponent bias that fails to capture the full complexity of realistic preference structures. In this work, we introduce Multiplayer Nash Preference Optimization (MNPO), a novel framework that generalizes NLHF to the multiplayer regime. It formulates alignment as an n-player game, where each policy competes against a population of opponents while being regularized toward a reference model. Our framework establishes well-defined Nash equilibria in multiplayer settings and extends the concept of duality gap to quantify approximation quality. We demonstrate that MNPO inherits the equilibrium guarantees of two-player methods while enabling richer competitive dynamics and improved coverage of diverse preference structures. Through comprehensive empirical evaluation, we show that MNPO consistently outperforms existing NLHF baselines on instruction-following benchmarks, achieving superior alignment quality under heterogeneous annotator conditions and mixed-policy evaluation scenarios. Together, these results establish MNPO as a principled and scalable framework for aligning LLMs with complex, non-transitive human preferences. Code is available at https://github.com/smiles724/MNPO.

As Large Language Model (LLM) agents increasingly gain self-evolutionary capabilities to adapt and refine their strategies through real-world interaction, their long-term reliability becomes a critical concern. We identify the Alignment Tipping Process (ATP), a critical post-deployment risk unique to self-evolving LLM agents. Unlike training-time failures, ATP arises when continual interaction drives agents to abandon alignment constraints established during training in favor of reinforced, self-interested strategies. We formalize and analyze ATP through two complementary paradigms: Self-Interested Exploration, where repeated high-reward deviations induce individual behavioral drift, and Imitative Strategy Diffusion, where deviant behaviors spread across multi-agent systems. Building on these paradigms, we construct controllable testbeds and benchmark Qwen3-8B and Llama-3.1-8B-Instruct. Our experiments show that alignment benefits erode rapidly under self-evolution, with initially aligned models converging toward unaligned states. In multi-agent settings, successful violations diffuse quickly, leading to collective misalignment. Moreover, current reinforcement learning-based alignment methods provide only fragile defenses against alignment tipping. Together, these findings demonstrate that alignment of LLM agents is not a static property but a fragile and dynamic one, vulnerable to feedback-driven decay during deployment. Our data and code are available at https://github.com/aiming-lab/ATP.

Like a criminal under investigation, Large Language Models (LLMs) might pretend to be aligned while evaluated and misbehave when they have a good opportunity. Can current interpretability methods catch these 'alignment fakers?' To answer this question, we introduce a benchmark that consists of 324 pairs of LLMs fine-tuned to select actions in role-play scenarios. One model in each pair is consistently benign (aligned). The other model misbehaves in scenarios where it is unlikely to be caught (alignment faking). The task is to identify the alignment faking model using only inputs where the two models behave identically. We test five detection strategies, one of which identifies 98% of alignment-fakers.

Large Language Models (LLMs) are increasingly deployed in socially complex, interaction-driven tasks, yet their ability to mirror human behavior in emotionally and strategically complex contexts remains underexplored. This study assesses the behavioral alignment of personality-prompted LLMs in adversarial dispute resolution by simulating multi-turn conflict dialogues that incorporate negotiation. Each LLM is guided by a matched Five-Factor personality profile to control for individual variation and enhance realism. We evaluate alignment across three dimensions: linguistic style, emotional expression (e.g., anger dynamics), and strategic behavior. GPT-4.1 achieves the closest alignment with humans in linguistic style and emotional dynamics, while Claude-3.7-Sonnet best reflects strategic behavior. Nonetheless, substantial alignment gaps persist. Our findings establish a benchmark for alignment between LLMs and humans in socially complex interactions, underscoring both the promise and the limitations of personality conditioning in dialogue modeling.

A common approach for aligning language models to human preferences is to first learn a reward model from preference data, and then use this reward model to update the language model. We study two closely related problems that arise in this approach. First, any monotone transformation of the reward model preserves preference ranking; is there a choice that is ``better'' than others? Second, we often wish to align language models to multiple properties: how should we combine multiple reward models? Using a probabilistic interpretation of the alignment procedure, we identify a natural choice for transformation for (the common case of) rewards learned from Bradley-Terry preference models. This derived transformation has two important properties. First, it emphasizes improving poorly-performing outputs, rather than outputs that already score well. This mitigates both underfitting (where some prompts are not improved) and reward hacking (where the model learns to exploit misspecification of the reward model). Second, it enables principled aggregation of rewards by linking summation to logical conjunction: the sum of transformed rewards corresponds to the probability that the output is ``good'' in all measured properties, in a sense we make precise. Experiments aligning language models to be both helpful and harmless using RLHF show substantial improvements over the baseline (non-transformed) approach.

Recent research shows that Preference Alignment (PA) objectives act as divergence estimators between aligned (chosen) and unaligned (rejected) response distributions. In this work, we extend this divergence-based perspective to general alignment settings, such as reinforcement learning with verifiable rewards (RLVR), where only environmental rewards are available. Within this unified framework, we propose f-Group Relative Policy Optimization (f-GRPO), a class of on-policy reinforcement learning, and f-Hybrid Alignment Loss (f-HAL), a hybrid on/off policy objectives, for general LLM alignment based on variational representation of f-divergences. We provide theoretical guarantees that these classes of objectives improve the average reward after alignment. Empirically, we validate our framework on both RLVR (Math Reasoning) and PA tasks (Safety Alignment), demonstrating superior performance and flexibility compared to current methods.

Preference learning algorithms (e.g., RLHF and DPO) are frequently used to steer LLMs to produce generations that are more preferred by humans, but our understanding of their inner workings is still limited. In this work, we study the conventional wisdom that preference learning trains models to assign higher likelihoods to more preferred outputs than less preferred outputs, measured via ranking accuracy. Surprisingly, we find that most state-of-the-art preference-tuned models achieve a ranking accuracy of less than 60% on common preference datasets. We furthermore derive the idealized ranking accuracy that a preference-tuned LLM would achieve if it optimized the DPO or RLHF objective perfectly. We demonstrate that existing models exhibit a significant alignment gap -- i.e., a gap between the observed and idealized ranking accuracies. We attribute this discrepancy to the DPO objective, which is empirically and theoretically ill-suited to fix even mild ranking errors in the reference model, and derive a simple and efficient formula for quantifying the difficulty of learning a given preference datapoint. Finally, we demonstrate that ranking accuracy strongly correlates with the empirically popular win rate metric when the model is close to the reference model used in the objective, shedding further light on the differences between on-policy (e.g., RLHF) and off-policy (e.g., DPO) preference learning algorithms.

Aligning language models (LMs) with human preferences has become a key area of research, enabling these models to meet diverse user needs better. Inspired by weak-to-strong generalization, where a strong LM fine-tuned on labels generated by a weaker model can consistently outperform its weak supervisor, we extend this idea to model alignment. In this work, we observe that the alignment behavior in weaker models can be effectively transferred to stronger models and even exhibit an amplification effect. Based on this insight, we propose a method called Weak-to-Strong Preference Optimization (WSPO), which achieves strong model alignment by learning the distribution differences before and after the alignment of the weak model. Experiments demonstrate that WSPO delivers outstanding performance, improving the win rate of Qwen2-7B-Instruct on Arena-Hard from 39.70 to 49.60, achieving a remarkable 47.04 length-controlled win rate on AlpacaEval 2, and scoring 7.33 on MT-bench. Our results suggest that using the weak model to elicit a strong model with a high alignment ability is feasible.

Foundation models, specifically Large Language Models (LLMs), have lately gained wide-spread attention and adoption. Reinforcement Learning with Human Feedback (RLHF) involves training a reward model to capture desired behaviors, which is then used to align LLM's. These reward models are additionally used at inference-time to estimate LLM responses' adherence to those desired behaviors. However, there is little work measuring how robust these reward models are to distribution shifts. In this work, we evaluate how reward model performance - measured via accuracy and calibration (i.e. alignment between accuracy and confidence) - is affected by distribution shift. We show novel calibration patterns and accuracy drops due to OOD prompts and responses, and that the reward model is more sensitive to shifts in responses than prompts. Additionally, we adapt an OOD detection technique commonly used in classification to the reward model setting to detect these distribution shifts in prompts and responses.

While astonishingly capable, large Language Models (LLM) can sometimes produce outputs that deviate from human expectations. Such deviations necessitate an alignment phase to prevent disseminating untruthful, toxic, or biased information. Traditional alignment methods based on reinforcement learning often struggle with the identified instability, whereas preference optimization methods are limited by their overfitting to pre-collected hard-label datasets. In this paper, we propose a novel LLM alignment framework named iREPO, which utilizes implicit Reward pairwise difference regression for Empirical Preference Optimization. Particularly, iREPO employs self-generated datasets labelled by empirical human (or AI annotator) preference to iteratively refine the aligned policy through a novel regression-based loss function. Furthermore, we introduce an innovative algorithm backed by theoretical guarantees for achieving optimal results under ideal assumptions and providing a practical performance-gap result without such assumptions. Experimental results with Phi-2 and Mistral-7B demonstrate that iREPO effectively achieves self-alignment using soft-label, self-generated responses and the logit of empirical AI annotators. Furthermore, our approach surpasses preference optimization baselines in evaluations using the Language Model Evaluation Harness and Multi-turn benchmarks.

Large language models (LLMs) are increasingly used for tasks that require complex reasoning. Most benchmarks focus on final outcomes but overlook the intermediate reasoning steps - such as planning, revision, and decision making under resource constraints. We argue that measuring these internal processes is essential for understanding model behavior and improving reliability. We propose using strategic games as a natural evaluation environment: closed, rule-based systems with clear states, limited resources, and automatic feedback. We introduce a framework that evaluates LLMs along three core dimensions: planning, revision, and resource-constrained decision making. To operationalize this, we define metrics beyond win rate, including overcorrection risk rate, correction success rate, improvement slope, and over-budget ratio. In 4320 adversarial rounds across 12 leading models, ChatGPT-o3-mini achieves the top composite score, with a win rate of 74.7 percent, a correction success rate of 78.6 percent, and an improvement slope of 0.041. By contrast, Qwen-Plus, despite an overcorrection risk rate of 81.6 percent, wins only 25.6 percent of its matches - primarily due to excessive resource use. We also observe a negative correlation between overcorrection risk rate and correction success rate (Pearson r = -0.51, p = 0.093), suggesting that more frequent edits do not always improve outcomes. Our findings highlight the value of assessing not only what LLMs decide but how they arrive at those decisions

Big models, exemplified by Large Language Models (LLMs), are models typically pre-trained on massive data and comprised of enormous parameters, which not only obtain significantly improved performance across diverse tasks but also present emergent capabilities absent in smaller models. However, the growing intertwining of big models with everyday human lives poses potential risks and might cause serious social harm. Therefore, many efforts have been made to align LLMs with humans to make them better follow user instructions and satisfy human preferences. Nevertheless, `what to align with' has not been fully discussed, and inappropriate alignment goals might even backfire. In this paper, we conduct a comprehensive survey of different alignment goals in existing work and trace their evolution paths to help identify the most essential goal. Particularly, we investigate related works from two perspectives: the definition of alignment goals and alignment evaluation. Our analysis encompasses three distinct levels of alignment goals and reveals a goal transformation from fundamental abilities to value orientation, indicating the potential of intrinsic human values as the alignment goal for enhanced LLMs. Based on such results, we further discuss the challenges of achieving such intrinsic value alignment and provide a collection of available resources for future research on the alignment of big models.

Recent years have witnessed remarkable progress made in large language models (LLMs). Such advancements, while garnering significant attention, have concurrently elicited various concerns. The potential of these models is undeniably vast; however, they may yield texts that are imprecise, misleading, or even detrimental. Consequently, it becomes paramount to employ alignment techniques to ensure these models to exhibit behaviors consistent with human values. This survey endeavors to furnish an extensive exploration of alignment methodologies designed for LLMs, in conjunction with the extant capability research in this domain. Adopting the lens of AI alignment, we categorize the prevailing methods and emergent proposals for the alignment of LLMs into outer and inner alignment. We also probe into salient issues including the models' interpretability, and potential vulnerabilities to adversarial attacks. To assess LLM alignment, we present a wide variety of benchmarks and evaluation methodologies. After discussing the state of alignment research for LLMs, we finally cast a vision toward the future, contemplating the promising avenues of research that lie ahead. Our aspiration for this survey extends beyond merely spurring research interests in this realm. We also envision bridging the gap between the AI alignment research community and the researchers engrossed in the capability exploration of LLMs for both capable and safe LLMs.

Aligning large language models with human objectives is paramount, yet common approaches including RLHF suffer from unstable and resource-intensive training. In response to this challenge, we introduce ARGS, Alignment as Reward-Guided Search, a novel framework that integrates alignment into the decoding process, eliminating the need for expensive RL training. By adjusting the model's probabilistic predictions using a reward signal, ARGS generates texts with semantic diversity while being aligned with human preferences, offering a promising and flexible solution for aligning language models. Notably, ARGS demonstrates consistent enhancements in average reward compared to baselines across diverse alignment tasks and various model dimensions. For example, under the same greedy-based decoding strategy, our method improves the average reward by 19.56% relative to the baseline and secures a preference or tie score of 64.33% in GPT-4 evaluation. We believe that our framework, emphasizing decoding-time alignment, paves the way for more responsive language models in the future. Code is publicly available at: https://github.com/deeplearning-wisc/args.

Alignment in artificial intelligence pursues the consistency between model responses and human preferences as well as values. In practice, the multifaceted nature of human preferences inadvertently introduces what is known as the "alignment tax" -a compromise where enhancements in alignment within one objective (e.g.,harmlessness) can diminish performance in others (e.g.,helpfulness). However, existing alignment techniques are mostly unidirectional, leading to suboptimal trade-offs and poor flexibility over various objectives. To navigate this challenge, we argue the prominence of grounding LLMs with evident preferences. We introduce controllable preference optimization (CPO), which explicitly specifies preference scores for different objectives, thereby guiding the model to generate responses that meet the requirements. Our experimental analysis reveals that the aligned models can provide responses that match various preferences among the "3H" (helpfulness, honesty, harmlessness) desiderata. Furthermore, by introducing diverse data and alignment goals, we surpass baseline methods in aligning with single objectives, hence mitigating the impact of the alignment tax and achieving improvements in multi-objective alignment.

Preference datasets are essential for incorporating human preferences into pre-trained language models, playing a key role in the success of Reinforcement Learning from Human Feedback. However, these datasets often demonstrate conflicting alignment objectives, leading to increased vulnerability to jailbreak attacks and challenges in adapting downstream tasks to prioritize specific alignment objectives without negatively impacting others. In this work, we introduce a novel statistical metric, Alignment Dimension Conflict, to quantify the degree of conflict within preference datasets. We then present Hummer and its fine-grained variant, Hummer-F, as innovative pairwise preference datasets with reduced-conflict alignment objectives. Hummer is built based on UltraFeedback and is enhanced by AI feedback from GPT-4, marking as the first preference dataset aimed at reducing the competition between alignment objectives. Furthermore, we develop reward models, HummerRM and HummerRM-F, which employ a hybrid sampling approach to balance diverse alignment objectives effectively. This sampling method positions HummerRM as an ideal model for domain-specific further fine-tuning and reducing vulnerabilities to attacks.

The widespread applicability and increasing omnipresence of LLMs have instigated a need to align LLM responses to user and stakeholder preferences. Many preference optimization approaches have been proposed that fine-tune LLM parameters to achieve good alignment. However, such parameter tuning is known to interfere with model performance on many tasks. Moreover, keeping up with shifting user preferences is tricky in such a situation. Decoding-time alignment with reward model guidance solves these issues at the cost of increased inference time. However, most of such methods fail to strike the right balance between exploration and exploitation of reward -- often due to the conflated formulation of these two aspects - to give well-aligned responses. To remedy this we decouple these two aspects and implement them in an evolutionary fashion: exploration is enforced by decoding from mutated instructions and exploitation is represented as the periodic replacement of poorly-rewarded generations with well-rewarded ones. Empirical evidences indicate that this strategy outperforms many preference optimization and decode-time alignment approaches on two widely accepted alignment benchmarks AlpacaEval 2 and MT-Bench. Our implementation will be available at: https://darwin-alignment.github.io.

Reinforcement learning from human feedback (RLHF) has demonstrated remarkable effectiveness in aligning large language models (LLMs) with human preferences. Many existing alignment approaches rely on the Bradley-Terry (BT) model assumption, which assumes the existence of a ground-truth reward for each prompt-response pair. However, this assumption can be overly restrictive when modeling complex human preferences. In this paper, we drop the BT model assumption and study LLM alignment under general preferences, formulated as a two-player game. Drawing on theoretical insights from learning in games, we integrate optimistic online mirror descent into our alignment framework to approximate the Nash policy. Theoretically, we demonstrate that our approach achieves an O(T^{-1}) bound on the duality gap, improving upon the previous O(T^{-1/2}) result. More importantly, we implement our method and show through experiments that it outperforms state-of-the-art RLHF algorithms across multiple representative benchmarks.

A key concern with the concept of "alignment" is the implicit question of "alignment to what?". AI systems are increasingly used across the world, yet safety alignment is often focused on homogeneous monolingual settings. Additionally, preference training and safety measures often overfit to harms common in Western-centric datasets. Here, we explore the viability of different alignment approaches when balancing dual objectives: addressing and optimizing for a non-homogeneous set of languages and cultural preferences while minimizing both global and local harms. We collect the first set of human annotated red-teaming prompts in different languages distinguishing between global and local harm, which serve as a laboratory for understanding the reliability of alignment techniques when faced with preference distributions that are non-stationary across geographies and languages. While this setting is seldom covered by the literature to date, which primarily centers on English harm mitigation, it captures real-world interactions with AI systems around the world. We establish a new precedent for state-of-the-art alignment techniques across 6 languages with minimal degradation in general performance. Our work provides important insights into cross-lingual transfer and novel optimization approaches to safeguard AI systems designed to serve global populations.

Before being deployed for user-facing applications, developers align Large Language Models (LLMs) to user preferences through a variety of procedures, such as Reinforcement Learning From Human Feedback (RLHF) and Direct Preference Optimization (DPO). Current evaluations of these procedures focus on benchmarks of instruction following, reasoning, and truthfulness. However, human preferences are not universal, and aligning to specific preference sets may have unintended effects. We explore how alignment impacts performance along three axes of global representation: English dialects, multilingualism, and opinions from and about countries worldwide. Our results show that current alignment procedures create disparities between English dialects and global opinions. We find alignment improves capabilities in several languages. We conclude by discussing design decisions that led to these unintended impacts and recommendations for more equitable preference tuning.

Direct alignment methods are increasingly used to align large language models (LLMs) with human preferences. However, many real-world alignment problems involve multiple conflicting objectives, where naive aggregation of preferences can lead to unstable training and poor trade-offs. In particular, weighted loss methods may fail to identify update directions that simultaneously improve all objectives, and existing multi-objective approaches often rely on explicit reward models, introducing additional complexity and distorting user-specified preferences. The contributions of this paper are two-fold. First, we propose a Reward-free Alignment framework for Conflicted Objectives (RACO) that directly leverages pairwise preference data and resolves gradient conflicts via a novel clipped variant of conflict-averse gradient descent. We provide convergence guarantees to Pareto-critical points that respect user-specified objective weights, and further show that clipping can strictly improve convergence rate in the two-objective setting. Second, we improve our method using some heuristics and conduct experiments to demonstrate the compatibility of the proposed framework for LLM alignment. Both qualitative and quantitative evaluations on multi-objective summarization and safety alignment tasks across multiple LLM families (Qwen 3, Llama 3, Gemma 3) show that our method consistently achieves better Pareto trade-offs compared to existing multi-objective alignment baselines.

As humans, we consistently engage in interactions with our peers and receive feedback in the form of natural language. This language feedback allows us to reflect on our actions, maintain appropriate behavior, and rectify our errors. The question arises naturally: can we use language feedback to align large language models (LLMs)? In contrast to previous research that aligns LLMs with reward or preference data, we present the first systematic exploration of alignment through the lens of language feedback (i.e., judgment). We commence with an in-depth investigation of potential methods that can be adapted for aligning LLMs with judgments, revealing that these methods are unable to fully capitalize on the judgments. To facilitate more effective utilization of judgments, we propose a novel framework, Contrastive Unlikelihood Training (CUT), that allows for fine-grained inappropriate content detection and correction based on judgments. Our offline alignment results show that, with merely 1317 off-the-shelf judgment data, CUT (LLaMA2-13b) can beat the 175B DaVinci003 and surpass the best baseline by 52.34 points on AlpacaEval. The online alignment results demonstrate that CUT can align LLMs (LLaMA2-chat-13b) in an iterative fashion using model-specific judgment data, with a steady performance improvement from 81.09 to 91.36 points on AlpacaEval. Our analysis further suggests that judgments exhibit greater potential than rewards for LLM alignment and warrant future research.

Alignment methodologies have emerged as a critical pathway for enhancing language model alignment capabilities. While SFT (supervised fine-tuning) accelerates convergence through direct token-level loss intervention, its efficacy is constrained by offline policy trajectory. In contrast, RL(reinforcement learning) facilitates exploratory policy optimization, but suffers from low sample efficiency and stringent dependency on high-quality base models. To address these dual challenges, we propose GRAO (Group Relative Alignment Optimization), a unified framework that synergizes the respective strengths of SFT and RL through three key innovations: 1) A multi-sample generation strategy enabling comparative quality assessment via reward feedback; 2) A novel Group Direct Alignment Loss formulation leveraging intra-group relative advantage weighting; 3) Reference-aware parameter updates guided by pairwise preference dynamics. Our theoretical analysis establishes GRAO's convergence guarantees and sample efficiency advantages over conventional approaches. Comprehensive evaluations across complex human alignment tasks demonstrate GRAO's superior performance, achieving 57.70\%,17.65\% 7.95\% and 5.18\% relative improvements over SFT, DPO, PPO and GRPO baselines respectively. This work provides both a theoretically grounded alignment framework and empirical evidence for efficient capability evolution in language models.

Effectively aligning Large Language Models (LLMs) with human-centric values while preventing the degradation of abilities acquired through Pre-training and Supervised Fine-tuning (SFT) poses a central challenge in Reinforcement Learning from Human Feedback (RLHF). In this paper, we first discover that interpolating RLHF and SFT model parameters can adjust the trade-off between human preference and basic capabilities, thereby reducing the alignment tax at the cost of alignment reward. Inspired by this, we propose integrating the RL policy and SFT models at each optimization step in RLHF to continuously regulate the training direction, introducing the Online Merging Optimizer. Specifically, we merge gradients with the parameter differences between SFT and pretrained models, effectively steering the gradient towards maximizing rewards in the direction of SFT optimization. We demonstrate that our optimizer works well with different LLM families, such as Qwen and LLaMA, across various model sizes ranging from 1.8B to 8B, various RLHF algorithms like DPO and KTO, and existing model merging methods. It significantly enhances alignment reward while mitigating alignment tax, achieving higher overall performance across 14 benchmarks.

Current LLM alignment techniques use pairwise human preferences at a sample level, and as such, they do not imply an alignment on the distributional level. We propose in this paper Alignment via Optimal Transport (AOT), a novel method for distributional preference alignment of LLMs. AOT aligns LLMs on unpaired preference data by making the reward distribution of the positive samples stochastically dominant in the first order on the distribution of negative samples. We introduce a convex relaxation of this first-order stochastic dominance and cast it as an optimal transport problem with a smooth and convex cost. Thanks to the one-dimensional nature of the resulting optimal transport problem and the convexity of the cost, it has a closed-form solution via sorting on empirical measures. We fine-tune LLMs with this AOT objective, which enables alignment by penalizing the violation of the stochastic dominance of the reward distribution of the positive samples on the reward distribution of the negative samples. We analyze the sample complexity of AOT by considering the dual of the OT problem and show that it converges at the parametric rate. Empirically, we show on a diverse set of alignment datasets and LLMs that AOT leads to state-of-the-art models in the 7B family of models when evaluated with Open LLM Benchmarks and AlpacaEval.

Reinforcement Learning from Human Feedback (RLHF) is widely used for aligning large language models, yet practitioners face a persistent puzzle: improving safety often reduces fairness, scaling to diverse populations becomes computationally intractable, and making systems robust often amplifies majority biases. We formalize this tension as the Alignment Trilemma: no RLHF system can simultaneously achieve (i) epsilon-representativeness across diverse human values, (ii) polynomial tractability in sample and compute complexity, and (iii) delta-robustness against adversarial perturbations and distribution shift. Through a complexity-theoretic analysis integrating statistical learning theory and robust optimization, we prove that achieving both representativeness (epsilon <= 0.01) and robustness (delta <= 0.001) for global-scale populations requires Omega(2^{d_context}) operations, which is super-polynomial in the context dimensionality. We show that current RLHF implementations resolve this trilemma by sacrificing representativeness: they collect only 10^3--10^4 samples from homogeneous annotator pools while 10^7--10^8 samples are needed for true global representation. Our framework provides a unified explanation for documented RLHF pathologies including preference collapse, sycophancy, and systematic bias amplification. We conclude with concrete directions for navigating these fundamental trade-offs through strategic relaxations of alignment requirements.

Generative Reward Models (GenRMs) and LLM-as-a-Judge exhibit deceptive alignment by producing correct judgments for incorrect reasons, as they are trained and evaluated to prioritize Outcome Accuracy, which undermines their ability to generalize during RLHF. We introduce Rationale Consistency, a fine-grained metric that quantifies the alignment between the model's reasoning process and human judgment. Our evaluation of frontier models reveals that rationale consistency effectively discriminates among state-of-the-art models and detects deceptive alignment, while outcome accuracy falls short in both respects. To mitigate this gap, we introduce a hybrid signal that combines rationale consistency with outcome accuracy for GenRM training. Our training method achieves state-of-the-art performance on RM-Bench (87.1%) and JudgeBench (82%), surpassing outcome-only baselines by an average of 5%. Using RM during RLHF, our method effectively improves performance as demonstrated on Arena Hard v2, notably yielding a 7% improvement in creative writing tasks. Further analysis confirms that our method escapes the deceptive alignment trap, effectively reversing the decline in rationale consistency observed in outcome-only training.

Alignment is no longer a luxury, it is a necessity. As large language models (LLMs) enter high-stakes domains like education, healthcare, governance, and law, their behavior must reliably reflect human-aligned values and safety constraints. Yet current evaluations rely heavily on behavioral proxies such as refusal rates, G-Eval scores, and toxicity classifiers, all of which have critical blind spots. Aligned models are often vulnerable to jailbreaking, stochasticity of generation, and alignment faking. To address this issue, we introduce the Alignment Quality Index (AQI). This novel geometric and prompt-invariant metric empirically assesses LLM alignment by analyzing the separation of safe and unsafe activations in latent space. By combining measures such as the Davies-Bouldin Score (DBS), Dunn Index (DI), Xie-Beni Index (XBI), and Calinski-Harabasz Index (CHI) across various formulations, AQI captures clustering quality to detect hidden misalignments and jailbreak risks, even when outputs appear compliant. AQI also serves as an early warning signal for alignment faking, offering a robust, decoding invariant tool for behavior agnostic safety auditing. Additionally, we propose the LITMUS dataset to facilitate robust evaluation under these challenging conditions. Empirical tests on LITMUS across different models trained under DPO, GRPO, and RLHF conditions demonstrate AQI's correlation with external judges and ability to reveal vulnerabilities missed by refusal metrics. We make our implementation publicly available to foster future research in this area.

As agents based on large language models are increasingly deployed to long-horizon tasks, maintaining their alignment with stakeholder preferences becomes critical. Effective alignment in such settings requires reward models that are interpretable so that stakeholders can understand and audit model objectives. Moreover, reward models must be capable of steering agents at interaction time, allowing preference shifts to be incorporated without retraining. We introduce ARCANE, a framework that frames alignment as a multi-agent collaboration problem that dynamically represents stakeholder preferences as natural-language rubrics: weighted sets of verifiable criteria that can be generated on-the-fly from task context. Inspired by utility theory, we formulate rubric learning as a reconstruction problem and apply a regularized Group-Sequence Policy Optimization (GSPO) procedure that balances interpretability, faithfulness, and computational efficiency. Using a corpus of 219 labeled rubrics derived from the GDPVal benchmark, we evaluate ARCANE on challenging tasks requiring multi-step reasoning and tool use. The learned rubrics produce compact, legible evaluations and enable configurable trade-offs (e.g., correctness vs. conciseness) without retraining. Our results show that rubric-based reward models offer a promising path toward interpretable, test-time adaptive alignment for complex, long-horizon AI systems.

Language model alignment research often attempts to ensure that models are not only helpful and harmless, but also truthful and unbiased. However, optimizing these objectives simultaneously can obscure how improving one aspect might impact the others. In this work, we focus on analyzing the relationship between two concepts essential in both language model alignment and political science: truthfulness and political bias. We train reward models on various popular truthfulness datasets and subsequently evaluate their political bias. Our findings reveal that optimizing reward models for truthfulness on these datasets tends to result in a left-leaning political bias. We also find that existing open-source reward models (i.e. those trained on standard human preference datasets) already show a similar bias and that the bias is larger for larger models. These results raise important questions about both the datasets used to represent truthfulness and what language models capture about the relationship between truth and politics.

AI Alignment is often presented as an interaction between a single designer and an artificial agent in which the designer attempts to ensure the agent's behavior is consistent with its purpose, and risks arise solely because of conflicts caused by inadvertent misalignment between the utility function intended by the designer and the resulting internal utility function of the agent. With the advent of agents instantiated with large-language models (LLMs), which are typically pre-trained, we argue this does not capture the essential aspects of AI safety because in the real world there is not a one-to-one correspondence between designer and agent, and the many agents, both artificial and human, have heterogeneous values. Therefore, there is an economic aspect to AI safety and the principal-agent problem is likely to arise. In a principal-agent problem conflict arises because of information asymmetry together with inherent misalignment between the utility of the agent and its principal, and this inherent misalignment cannot be overcome by coercing the agent into adopting a desired utility function through training. We argue the assumptions underlying principal-agent problems are crucial to capturing the essence of safety problems involving pre-trained AI models in real-world situations. Taking an empirical approach to AI safety, we investigate how GPT models respond in principal-agent conflicts. We find that agents based on both GPT-3.5 and GPT-4 override their principal's objectives in a simple online shopping task, showing clear evidence of principal-agent conflict. Surprisingly, the earlier GPT-3.5 model exhibits more nuanced behaviour in response to changes in information asymmetry, whereas the later GPT-4 model is more rigid in adhering to its prior alignment. Our results highlight the importance of incorporating principles from economics into the alignment process.

Pretraining corpora contain extensive discourse about AI systems, yet the causal influence of this discourse on downstream alignment remains poorly understood. If prevailing descriptions of AI behaviour are predominantly negative, LLMs may internalise corresponding behavioural priors, giving rise to self-fulfilling misalignment. This paper provides the first controlled study of this hypothesis by pretraining 6.9B-parameter LLMs with varying amounts of (mis)alignment discourse. We find that discussion of AI contributes to misalignment. Upsampling synthetic training documents about AI misalignment leads to a notable increase in misaligned behaviour. Conversely, upsampling documents about aligned behaviour reduces misalignment scores from 45% to 9%. We consider this evidence of self-fulfilling alignment. These effects are dampened, but persist through post-training. Our findings establish the study of how pretraining data shapes alignment priors, or alignment pretraining, as a complement to post-training. We recommend practitioners pretrain for alignment as well as capabilities. Our models and datasets are available at alignmentpretraining.ai

Radiologists play a crucial role by translating medical images into medical reports. However, the field faces staffing shortages and increasing workloads. While automated approaches using vision-language models (VLMs) show promise as assistants, they require exceptionally high accuracy. Most current VLMs in radiology rely solely on supervised fine-tuning (SFT). Meanwhile, in the general domain, additional preference fine-tuning has become standard practice. The challenge in radiology lies in the prohibitive cost of obtaining radiologist feedback. We propose a scalable automated preference alignment technique for VLMs in radiology, focusing on chest X-ray (CXR) report generation. Our method leverages publicly available datasets with an LLM-as-a-Judge mechanism, eliminating the need for additional expert radiologist feedback. We evaluate and benchmark five direct alignment algorithms (DAAs). Our results show up to a 57.4% improvement in average GREEN scores, a LLM-based metric for evaluating CXR reports, and a 9.2% increase in an average across six metrics (domain specific and general), compared to the SFT baseline. We study reward overoptimization via length exploitation, with reports lengthening by up to 3.2x. To assess a potential alignment tax, we benchmark on six additional diverse tasks, finding no significant degradations. A reader study involving four board-certified radiologists indicates win rates of up to 0.62 over the SFT baseline, while significantly penalizing verbosity. Our analysis provides actionable insights for the development of VLMs in high-stakes fields like radiology.

The deployment of large language models (LLMs) raises significant ethical and safety concerns. While LLM alignment techniques are adopted to improve model safety and trustworthiness, adversaries can exploit these techniques to undermine safety for malicious purposes, resulting in misalignment. Misaligned LLMs may be published on open platforms to magnify harm. To address this, additional safety alignment, referred to as realignment, is necessary before deploying untrusted third-party LLMs. This study explores the efficacy of fine-tuning methods in terms of misalignment, realignment, and the effects of their interplay. By evaluating four Supervised Fine-Tuning (SFT) and two Preference Fine-Tuning (PFT) methods across four popular safety-aligned LLMs, we reveal a mechanism asymmetry between attack and defense. While Odds Ratio Preference Optimization (ORPO) is most effective for misalignment, Direct Preference Optimization (DPO) excels in realignment, albeit at the expense of model utility. Additionally, we identify model-specific resistance, residual effects of multi-round adversarial dynamics, and other noteworthy findings. These findings highlight the need for robust safeguards and customized safety alignment strategies to mitigate potential risks in the deployment of LLMs. Our code is available at https://github.com/zhangrui4041/The-Art-of-Mis-alignment.

Aligning language models with preferences can be posed as approximating a target distribution representing some desired behavior. Existing approaches differ both in the functional form of the target distribution and the algorithm used to approximate it. For instance, Reinforcement Learning from Human Feedback (RLHF) corresponds to minimizing a reverse KL from an implicit target distribution arising from a KL penalty in the objective. On the other hand, Generative Distributional Control (GDC) has an explicit target distribution and minimizes a forward KL from it using the Distributional Policy Gradient (DPG) algorithm. In this paper, we propose a new approach, f-DPG, which allows the use of any f-divergence to approximate any target distribution that can be evaluated. f-DPG unifies both frameworks (RLHF, GDC) and the approximation methods (DPG, RL with KL penalties). We show the practical benefits of various choices of divergence objectives and demonstrate that there is no universally optimal objective but that different divergences present different alignment and diversity trade-offs. We show that Jensen-Shannon divergence strikes a good balance between these objectives, and frequently outperforms forward KL divergence by a wide margin, leading to significant improvements over prior work. These distinguishing characteristics between divergences persist as the model size increases, highlighting the importance of selecting appropriate divergence objectives.

Ensuring alignment, which refers to making models behave in accordance with human intentions [1,2], has become a critical task before deploying large language models (LLMs) in real-world applications. For instance, OpenAI devoted six months to iteratively aligning GPT-4 before its release [3]. However, a major challenge faced by practitioners is the lack of clear guidance on evaluating whether LLM outputs align with social norms, values, and regulations. This obstacle hinders systematic iteration and deployment of LLMs. To address this issue, this paper presents a comprehensive survey of key dimensions that are crucial to consider when assessing LLM trustworthiness. The survey covers seven major categories of LLM trustworthiness: reliability, safety, fairness, resistance to misuse, explainability and reasoning, adherence to social norms, and robustness. Each major category is further divided into several sub-categories, resulting in a total of 29 sub-categories. Additionally, a subset of 8 sub-categories is selected for further investigation, where corresponding measurement studies are designed and conducted on several widely-used LLMs. The measurement results indicate that, in general, more aligned models tend to perform better in terms of overall trustworthiness. However, the effectiveness of alignment varies across the different trustworthiness categories considered. This highlights the importance of conducting more fine-grained analyses, testing, and making continuous improvements on LLM alignment. By shedding light on these key dimensions of LLM trustworthiness, this paper aims to provide valuable insights and guidance to practitioners in the field. Understanding and addressing these concerns will be crucial in achieving reliable and ethically sound deployment of LLMs in various applications.

The recent advancement of large language models (LLMs) has been achieved through a combo of instruction tuning and human alignment. However, building manually crafted instruction datasets and performing human alignment become the bottleneck for scaling the development of LLMs. In this paper, we exploit the idea of leveraging AI models in lieu of humans as the teacher to train student LLMs. Our method is inspired by how human students refine their writing skills by following the rubrics and learning from the revisions offered by their tutors. Specifically, we employ a teacher LLM to create a curriculum for instruction tuning of the student LLM, namely Curriculum Instruction TunING (CITING). It encompasses two main steps: (1) the teacher LLM crafts the rubrics for evaluating the answers corresponding to various types of questions, and (2) the student LLM learns to follow the rubrics and perform self-correction from the revision made by the teacher. We further iteratively carry out it to embody the procedure of CITING. We compare CITING to a series of state-of-the-art baselines on four datasets. Our method demonstrates strong improvement in terms of articulate, in-depth, and comprehensive by GPT-4 evaluation. Specifically, it achieves an average winning rate of 79.4% over SFT, 73.4% over RLHF, 78.1% over RRHF, and 76.3% over RAFT, respectively.

Aligning language models with human preferences is crucial for reducing errors and biases in these models. Alignment techniques, such as reinforcement learning from human feedback (RLHF), are typically cast as optimizing a tradeoff between human preference rewards and a proximity regularization term that encourages staying close to the unaligned model. Selecting an appropriate level of regularization is critical: insufficient regularization can lead to reduced model capabilities due to reward hacking, whereas excessive regularization hinders alignment. Traditional methods for finding the optimal regularization level require retraining multiple models with varying regularization strengths. This process, however, is resource-intensive, especially for large models. To address this challenge, we propose decoding-time realignment (DeRa), a simple method to explore and evaluate different regularization strengths in aligned models without retraining. DeRa enables control over the degree of alignment, allowing users to smoothly transition between unaligned and aligned models. It also enhances the efficiency of hyperparameter tuning by enabling the identification of effective regularization strengths using a validation dataset.

Alignment with human preference is a desired property of large language models (LLMs). Currently, the main alignment approach is based on reinforcement learning from human feedback (RLHF). Despite the effectiveness of RLHF, it is intricate to implement and train, thus recent studies explore how to develop alternative alignment approaches based on supervised fine-tuning (SFT). A major limitation of SFT is that it essentially does imitation learning, which cannot fully understand what are the expected behaviors. To address this issue, we propose an improved alignment approach named FIGA. Different from prior methods, we incorporate fine-grained (i.e., token or phrase level) quality signals that are derived by contrasting good and bad responses. Our approach has made two major contributions. Firstly, we curate a refined alignment dataset that pairs initial responses and the corresponding revised ones. Secondly, we devise a new loss function can leverage fine-grained quality signals to instruct the learning of LLMs for alignment. Extensive experiments have demonstrated the effectiveness of our approaches by comparing a number of competitive baselines.

Aligning large language models (LLMs) to human values has become increasingly important as it enables sophisticated steering of LLMs, e.g., making them follow given instructions while keeping them less toxic. However, it requires a significant amount of human demonstrations and feedback. Recently, open-sourced models have attempted to replicate the alignment learning process by distilling data from already aligned LLMs like InstructGPT or ChatGPT. While this process reduces human efforts, constructing these datasets has a heavy dependency on the teacher models. In this work, we propose a novel framework for alignment learning with almost no human labor and no dependency on pre-aligned LLMs. First, we perform reward modeling (RM) with synthetic feedback by contrasting responses from vanilla LLMs with various sizes and prompts. Then, we use the RM for simulating high-quality demonstrations to train a supervised policy and for further optimizing the model with reinforcement learning. Our resulting model, Aligned Language Model with Synthetic Training dataset (ALMoST), outperforms open-sourced models, including Alpaca, Dolly, and OpenAssistant, which are trained on the outputs of InstructGPT or human-annotated instructions. Our 7B-sized model outperforms the 12-13B models in the A/B tests using GPT-4 as the judge with about 75% winning rate on average.

Reinforcement learning from human feedback (RLHF) aligns large language models (LLMs) by encouraging their generations to have high rewards, using a reward model trained on human preferences. To prevent the forgetting of pre-trained knowledge, RLHF usually incorporates a KL regularization; this forces the policy to remain close to its supervised fine-tuned initialization, though it hinders the reward optimization. To tackle the trade-off between KL and reward, in this paper we introduce a novel alignment strategy named Weight Averaged Rewarded Policies (WARP). WARP merges policies in the weight space at three distinct stages. First, it uses the exponential moving average of the policy as a dynamic anchor in the KL regularization. Second, it applies spherical interpolation to merge independently fine-tuned policies into a new enhanced one. Third, it linearly interpolates between this merged model and the initialization, to recover features from pre-training. This procedure is then applied iteratively, with each iteration's final model used as an advanced initialization for the next, progressively refining the KL-reward Pareto front, achieving superior rewards at fixed KL. Experiments with GEMMA policies validate that WARP improves their quality and alignment, outperforming other open-source LLMs.

Aligning large language models (LLMs) with human intentions has become a critical task for safely deploying models in real-world systems. While existing alignment approaches have seen empirical success, theoretically understanding how these methods affect model behavior remains an open question. Our work provides an initial attempt to theoretically analyze the learning dynamics of human preference alignment. We formally show how the distribution of preference datasets influences the rate of model updates and provide rigorous guarantees on the training accuracy. Our theory also reveals an intricate phenomenon where the optimization is prone to prioritizing certain behaviors with higher preference distinguishability. We empirically validate our findings on contemporary LLMs and alignment tasks, reinforcing our theoretical insights and shedding light on considerations for future alignment approaches. Disclaimer: This paper contains potentially offensive text; reader discretion is advised.

The release of ChatGPT in November 2022 sparked an explosion of interest in post-training and an avalanche of new preference optimization (PO) methods. These methods claim superior alignment by virtue of better correspondence with human pairwise preferences, often measured by LLM judges. In this work, we attempt to answer the following question -- do LLM-judge preferences translate to progress on other, more concrete metrics for alignment, and if not, why not? We define a concrete metric for alignment, and introduce SOS-Bench, the largest standardized, reproducible LLM meta-benchmark to date. We find that (1) LLM-judgments do not correlate with concrete measures of safety, world knowledge, and instruction following; (2) LLM judges have powerful implicit biases, prioritizing style over factuality and safety; and (3) the supervised fine-tuning (SFT) stage of post-training, and not the PO stage, has the greatest impact on alignment, with data scaling and prompt diversity as the driving factors. Our codebase and complete results can be found at https://github.com/penfever/sos-bench.

Reinforcement Learning from Human Feedback (RLHF) has become the predominant approach for language model (LM) alignment. At its core, RLHF uses a margin-based loss for preference optimization, specifying ideal LM behavior only by the difference between preferred and dispreferred responses. In this paper, we identify a common pitfall of margin-based methods -- the under-specification of ideal LM behavior on preferred and dispreferred responses individually, which leads to two unintended consequences as the margin increases: (1) The probability of dispreferred (e.g., unsafe) responses may increase, resulting in potential safety alignment failures. (2) The probability of preferred responses may decrease, even when those responses are ideal. We demystify the reasons behind these problematic behaviors: margin-based losses couple the change in the preferred probability to the gradient of the dispreferred one, and vice versa, often preventing the preferred probability from increasing while the dispreferred one decreases, and thus causing a synchronized increase or decrease in both probabilities. We term this effect, inherent in margin-based objectives, gradient entanglement. Formally, we derive conditions for general margin-based alignment objectives under which gradient entanglement becomes concerning: the inner product of the gradients of preferred and dispreferred log-probabilities is large relative to the individual gradient norms. We theoretically investigate why such inner products can be large when aligning language models and empirically validate our findings. Empirical implications of our framework extend to explaining important differences in the training dynamics of various preference optimization algorithms, and suggesting potential algorithm designs to mitigate the under-specification issue of margin-based methods and thereby improving language model alignment.

Modern preference alignment techniques, such as Best-of-N (BoN) sampling, rely on reward models trained with pairwise comparison data. While effective at learning relative preferences, this paradigm fails to capture a signal of response acceptability, leaving systems vulnerable to selecting the least bad of many unacceptable options. This is particularly problematic for hard prompts, where the risk of such false acceptances increases with the number of samples. In this paper, we address this critical reliability gap by introducing a new data collection and modeling framework. By augmenting preference data with an outside option, inspired by discrete choice models, we train a reward model that can distinguish not just what is better, but what is good enough. We leverage this capability to create an adaptive inference strategy, best of mini-N in-loop, which partitions the generation budget into sequential loops with a calibrated, early-exit condition. Our experiments show that when tuned as an alignment guardrail, it reduces reliability failures by 70\%, and when tuned as an inference accelerator, it improves average inference speed by over 22\% in IMDB-sentiment setting. We thus provide a principled and flexible framework for practitioners to explicitly manage the trade-off between reliability and computational efficiency.

While reinforcement learning for large language model alignment has progressed rapidly in recent years, transferring these paradigms to high-stakes medical question answering reveals a fundamental paradigm mismatch. Reinforcement Learning from Human Feedback relies on preference annotations that are prohibitively expensive and often fail to reflect the absolute correctness of medical facts. Reinforcement Learning from Verifiable Rewards lacks effective automatic verifiers and struggles to handle complex clinical contexts. Meanwhile, medical alignment requires the simultaneous optimization of correctness, safety, and compliance, yet multi-objective heterogeneous reward signals are prone to scale mismatch and optimization conflicts.To address these challenges, we propose a robust medical alignment paradigm. We first construct a holistic multi-dimensional medical alignment matrix that decomposes alignment objectives into four categories: fundamental capabilities, expert knowledge, online feedback, and format specifications. Within each category, we establish a closed loop of where observable metrics inform attributable diagnosis, which in turn drives optimizable rewards, thereby providing fine-grained, high-resolution supervision signals for subsequent iterative optimization. To resolve gradient domination and optimization instability problem caused by heterogeneous signals, we further propose a unified optimization mechanism. This mechanism employs Reference-Frozen Normalization to align reward scales and implements a Tri-Factor Adaptive Dynamic Weighting strategy to achieve collaborative optimization that is weakness-oriented, risk-prioritized, and redundancy-reducing. Experimental results demonstrate the effectiveness of our proposed paradigm in real-world medical scenario evaluations, establishing a new paradigm for complex alignment in vertical domains.

This work studies the challenge of aligning large language models (LLMs) with offline preference data. We focus on alignment by Reinforcement Learning from Human Feedback (RLHF) in particular. While popular preference optimization methods exhibit good empirical performance in practice, they are not theoretically guaranteed to converge to the optimal policy and can provably fail when the data coverage is sparse by classical offline reinforcement learning (RL) results. On the other hand, a recent line of work has focused on theoretically motivated preference optimization methods with provable guarantees, but these are not computationally efficient for large-scale applications like LLM alignment. To bridge this gap, we propose SPAC, a new offline preference optimization method with self-play, inspired by the on-average pessimism technique from the offline RL literature, to be the first provable and scalable approach to LLM alignment. We both provide theoretical analysis for its convergence under single-policy concentrability for the general function approximation setting and demonstrate its competitive empirical performance for LLM alignment on a 7B Mistral model with Open LLM Leaderboard evaluations.

Recent advances in aligning Large Language Models with human preferences have benefited from larger reward models and better preference data. However, most of these methodologies rely on the accuracy of the reward model. The reward models used in Reinforcement Learning with Human Feedback (RLHF) are typically learned from small datasets using stochastic optimization algorithms, making them prone to high variability. We illustrate the inconsistencies between reward models empirically on numerous open-source datasets. We theoretically show that the fluctuation of the reward models can be detrimental to the alignment problem because the derived policies are more overfitted to the reward model and, hence, are riskier if the reward model itself is uncertain. We use concentration of measure to motivate an uncertainty-aware, conservative algorithm for policy optimization. We show that such policies are more risk-averse in the sense that they are more cautious of uncertain rewards. We theoretically prove that our proposed methodology has less risk than the vanilla method. We corroborate our theoretical results with experiments based on designing an ensemble of reward models. We use this ensemble of reward models to align a language model using our methodology and observe that our empirical findings match our theoretical predictions.

Aligning agent behaviors with diverse human preferences remains a challenging problem in reinforcement learning (RL), owing to the inherent abstractness and mutability of human preferences. To address these issues, we propose AlignDiff, a novel framework that leverages RL from Human Feedback (RLHF) to quantify human preferences, covering abstractness, and utilizes them to guide diffusion planning for zero-shot behavior customizing, covering mutability. AlignDiff can accurately match user-customized behaviors and efficiently switch from one to another. To build the framework, we first establish the multi-perspective human feedback datasets, which contain comparisons for the attributes of diverse behaviors, and then train an attribute strength model to predict quantified relative strengths. After relabeling behavioral datasets with relative strengths, we proceed to train an attribute-conditioned diffusion model, which serves as a planner with the attribute strength model as a director for preference aligning at the inference phase. We evaluate AlignDiff on various locomotion tasks and demonstrate its superior performance on preference matching, switching, and covering compared to other baselines. Its capability of completing unseen downstream tasks under human instructions also showcases the promising potential for human-AI collaboration. More visualization videos are released on https://aligndiff.github.io/.

Aligning large language models to human preferences is inherently multidimensional, yet most pipelines collapse heterogeneous signals into a single optimizeable objective. We seek to answer what it would take to simultaneously align a model across various domains spanning those with: verifiable rewards (mathematical accuracy), non-verifiable subjective preferences (human values), and complex interactive scenarios (multi-turn AI tutoring dialogues). Such multi-objective reinforcement learning setups are often plagued by the individual objectives being at odds with each other, resulting in inefficient training and little user control during inference. We propose a unified framework that: (i) standardizes {process reward model} (PRM) training across both verifiable and non-verifiable settings to better supervise models' chain-of-thought reasoning; (ii) performs {multi-objective alignment} by training the LLM with our Multi-Action-Head DPO (MAH-DPO) and a vectorized reward where the dimensions of the vector correspond to the various objectives instead of a single scalar; and (iii) demonstrates how such a system provides fine-grained inference-time user control. Experiments across math reasoning, value alignment, and multi-turn dialogue show that our framework improves performance across multiple objectives simultaneously, while minimizing cross-objective trade-offs and enabling flexible inference time user control. The code can be found at https://github.com/pearls-lab/multiobj-align.

As the capabilities of large language models (LLMs) have expanded dramatically, aligning these models with human values presents a significant challenge. Traditional alignment strategies rely heavily on human intervention, such as Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF), or on the self-alignment capacities of LLMs, which usually require a strong LLM's emergent ability to improve its original bad answer. To address these challenges, we propose a novel self-alignment method that utilizes a Chain of Thought (CoT) approach, termed AlignCoT. This method encompasses stages of Question Analysis, Answer Guidance, and Safe Answer production. It is designed to enable LLMs to generate high-quality, safe responses throughout various stages of their development. Furthermore, we introduce the Mixture of insighTful Experts (MoTE) architecture, which applies mixture of experts to enhance each component of the AlignCoT process, markedly increasing alignment efficiency. The MoTE approach not only outperforms existing methods in aligning LLMs with human values but also highlights the benefits of using self-generated data, revealing the dual benefits of improved alignment and training efficiency.

Due to the remarkable capabilities and growing impact of large language models (LLMs), they have been deeply integrated into many aspects of society. Thus, ensuring their alignment with human values and intentions has emerged as a critical challenge. This survey provides a comprehensive overview of practical alignment techniques, training protocols, and empirical findings in LLM alignment. We analyze the development of alignment methods across diverse paradigms, characterizing the fundamental trade-offs between core alignment objectives. Our analysis shows that while supervised fine-tuning enables basic instruction-following, preference-based methods offer more flexibility for aligning with nuanced human intent. We discuss state-of-the-art techniques, including Direct Preference Optimization (DPO), Constitutional AI, brain-inspired methods, and alignment uncertainty quantification (AUQ), highlighting their approaches to balancing quality and efficiency. We review existing evaluation frameworks and benchmarking datasets, emphasizing limitations such as reward misspecification, distributional robustness, and scalable oversight. We summarize strategies adopted by leading AI labs to illustrate the current state of practice. We conclude by outlining open problems in oversight, value pluralism, robustness, and continuous alignment. This survey aims to inform both researchers and practitioners navigating the evolving landscape of LLM alignment.

Large Language Models (LLMs) are increasingly tasked with creative generation, including the simulation of fictional characters. However, their ability to portray non-prosocial, antagonistic personas remains largely unexamined. We hypothesize that the safety alignment of modern LLMs creates a fundamental conflict with the task of authentically role-playing morally ambiguous or villainous characters. To investigate this, we introduce the Moral RolePlay benchmark, a new dataset featuring a four-level moral alignment scale and a balanced test set for rigorous evaluation. We task state-of-the-art LLMs with role-playing characters from moral paragons to pure villains. Our large-scale evaluation reveals a consistent, monotonic decline in role-playing fidelity as character morality decreases. We find that models struggle most with traits directly antithetical to safety principles, such as ``Deceitful'' and ``Manipulative'', often substituting nuanced malevolence with superficial aggression. Furthermore, we demonstrate that general chatbot proficiency is a poor predictor of villain role-playing ability, with highly safety-aligned models performing particularly poorly. Our work provides the first systematic evidence of this critical limitation, highlighting a key tension between model safety and creative fidelity. Our benchmark and findings pave the way for developing more nuanced, context-aware alignment methods.

Robust validation metrics remain essential in contemporary deep learning, not only to detect overfitting and poor generalization, but also to monitor training dynamics. In the supervised classification setting, we investigate whether interactions between training data and model weights can yield such a metric that both tracks generalization during training and attributes performance to individual training samples. We introduce Gradient-Weight Alignment (GWA), quantifying the coherence between per-sample gradients and model weights. We show that effective learning corresponds to coherent alignment, while misalignment indicates deteriorating generalization. GWA is efficiently computable during training and reflects both sample-specific contributions and dataset-wide learning dynamics. Extensive experiments show that GWA accurately predicts optimal early stopping, enables principled model comparisons, and identifies influential training samples, providing a validation-set-free approach for model analysis directly from the training data.

This study evaluates Artificial Intelligence (AI) agents for Dhumbal, a culturally significant multiplayer card game with imperfect information, through a systematic comparison of rule-based, search-based, and learning-based strategies. We formalize Dhumbal's mechanics and implement diverse agents, including heuristic approaches (Aggressive, Conservative, Balanced, Opportunistic), search-based methods such as Monte Carlo Tree Search (MCTS) and Information Set Monte Carlo Tree Search (ISMCTS), and reinforcement learning approaches including Deep Q-Network (DQN) and Proximal Policy Optimization (PPO), and a random baseline. Evaluation involves within-category tournaments followed by a cross-category championship. Performance is measured via win rate, economic outcome, Jhyap success, cards discarded per round, risk assessment, and decision efficiency. Statistical significance is assessed using Welch's t-test with Bonferroni correction, effect sizes via Cohen's d, and 95% confidence intervals (CI). Across 1024 simulated rounds, the rule-based Aggressive agent achieves the highest win rate (88.3%, 95% CI: [86.3, 90.3]), outperforming ISMCTS (9.0%) and PPO (1.5%) through effective exploitation of Jhyap declarations. The study contributes a reproducible AI framework, insights into heuristic efficacy under partial information, and open-source code, thereby advancing AI research and supporting digital preservation of cultural games.

Reinforcement learning from human feedback (RLHF) with reward models has advanced alignment of generative models to human aesthetic and perceptual preferences. However, jointly optimizing multiple rewards often incurs an alignment tax, improving one dimension while degrading others. To address this, we introduce two complementary methods: MapReduce LoRA and Reward-aware Token Embedding (RaTE). MapReduce LoRA trains preference-specific LoRA experts in parallel and iteratively merges them to refine a shared base model; RaTE learns reward-specific token embeddings that compose at inference for flexible preference control. Experiments on Text-to-Image generation (Stable Diffusion 3.5 Medium and FLUX.1-dev) show improvements of 36.1%, 4.6%, and 55.7%, and 32.7%, 4.3%, and 67.1% on GenEval, PickScore, and OCR, respectively. On Text-to-Video generation (HunyuanVideo), visual and motion quality improve by 48.1% and 90.0%, respectively. On the language task, Helpful Assistant, with Llama-2 7B, helpful and harmless improve by 43.4% and 136.7%, respectively. Our framework sets a new state-of-the-art multi-preference alignment recipe across modalities.

With the rapid development of large language models (LLMs), they are not only used as general-purpose AI assistants but are also customized through further fine-tuning to meet the requirements of different applications. A pivotal factor in the success of current LLMs is the alignment process. Current alignment methods, such as supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF), focus on training-time alignment and are often complex and cumbersome to implement. Therefore, we develop InferAligner, a novel inference-time alignment method that utilizes cross-model guidance for harmlessness alignment. InferAligner utilizes safety steering vectors extracted from safety-aligned model to modify the activations of the target model when responding to harmful inputs, thereby guiding the target model to provide harmless responses. Experimental results show that our method can be very effectively applied to domain-specific models in finance, medicine, and mathematics, as well as to multimodal large language models (MLLMs) such as LLaVA. It significantly diminishes the Attack Success Rate (ASR) of both harmful instructions and jailbreak attacks, while maintaining almost unchanged performance in downstream tasks.

Aligning Large Language Models (LLMs) with human preferences is crucial for their deployment in real-world applications. Recent advancements in Self-Rewarding Language Models suggest that an LLM can use its internal reward models (such as LLM-as-a-Judge) yuanself to generate preference data, improving alignment performance without costly human annotation. However, we find that different internal reward models within the same LLM often generate inconsistent preferences. This inconsistency raises concerns about the reliability of self-generated preference data, hinders overall alignment performance, and highlights the need for further research to ensure reliable and coherent alignment with human preferences. To address this limitation, we propose Self-Consistent Internal Rewards (SCIR), a novel framework designed to enhance consistency among internal reward models during training. In each training step, we collect preference predictions from multiple pre-defined internal reward models and enforce consistency and confidence through an inconsistency penalty mechanism, thereby improving the reliability of these internal reward models. We selectively use data with consistent predictions for preference optimization, ensuring the quality of the preference data. By employing self-consistent internal rewards, our method significantly improves the alignment performance and reward modeling capability of LLMs, outperforming baseline methods by a notable margin.

Aligning large language models (LLMs) with human preferences has been recognized as the key to improving LLMs' interaction quality. However, in this pluralistic world, human preferences can be diversified due to annotators' different tastes, which hinders the effectiveness of LLM alignment methods. This paper presents the first quantitative analysis of commonly used human feedback datasets to investigate the impact of diversified preferences on reward modeling. Our analysis reveals a correlation between the calibration performance of reward models (RMs) and the alignment performance of LLMs. We find that diversified preference data negatively affect the calibration performance of RMs on human-shared preferences, such as Harmless\&Helpful, thereby impairing the alignment performance of LLMs. To address the ineffectiveness, we propose a novel Multi-Objective Reward learning method (MORE) to enhance the calibration performance of RMs on shared preferences. We validate our findings by experiments on three models and five human preference datasets. Our method significantly improves the prediction calibration of RMs, leading to better alignment of the Alpaca-7B model with Harmless\&Helpful preferences. Furthermore, the connection between reward calibration and preference alignment performance suggests that calibration error can be adopted as a key metric for evaluating RMs. The open-source code and data are available at https://github.com/dunzeng/MORE.

Reinforcement Learning from Human Feedback (RLHF) remains indispensable for aligning large language models (LLMs) in subjective domains. To enhance robustness, recent work shifts toward Generative Reward Models (GenRMs) that generate rationales before predicting preferences. Yet in GenRM training and evaluation, practice remains outcome-label-only, leaving reasoning quality unchecked. We show that reasoning fidelity-the consistency between a GenRM's preference decision and reference decision rationales-is highly predictive of downstream RLHF outcomes, beyond standard label accuracy. Specifically, we repurpose existing reward-model benchmarks to compute Spurious Correctness (S-Corr)-the fraction of label-correct decisions with rationales misaligned with golden judgments. Our empirical evaluation reveals substantial S-Corr even for competitive GenRMs, and higher S-Corr is associated with policy degeneration under optimization. To improve fidelity, we propose Rationale-Centric Alignment, R-Align, which augments training with gold judgments and explicitly supervises rationale alignment. R-Align reduces S-Corr on RM benchmarks and yields consistent gains in actor performance across STEM, coding, instruction following, and general tasks.

The rapid progress in open-source large language models (LLMs) is significantly advancing AI development. Extensive efforts have been made before model release to align their behavior with human values, with the primary goal of ensuring their helpfulness and harmlessness. However, even carefully aligned models can be manipulated maliciously, leading to unintended behaviors, known as "jailbreaks". These jailbreaks are typically triggered by specific text inputs, often referred to as adversarial prompts. In this work, we propose the generation exploitation attack, an extremely simple approach that disrupts model alignment by only manipulating variations of decoding methods. By exploiting different generation strategies, including varying decoding hyper-parameters and sampling methods, we increase the misalignment rate from 0% to more than 95% across 11 language models including LLaMA2, Vicuna, Falcon, and MPT families, outperforming state-of-the-art attacks with 30times lower computational cost. Finally, we propose an effective alignment method that explores diverse generation strategies, which can reasonably reduce the misalignment rate under our attack. Altogether, our study underscores a major failure in current safety evaluation and alignment procedures for open-source LLMs, strongly advocating for more comprehensive red teaming and better alignment before releasing such models. Our code is available at https://github.com/Princeton-SysML/Jailbreak_LLM.

Post-training alignment of video generation models with human preferences is a critical goal. Developing effective Reward Models (RMs) for this process faces significant methodological hurdles. Current data collection paradigms, reliant on in-prompt pairwise annotations, suffer from labeling noise. Concurrently, the architectural design of VLM-based RMs, particularly their output mechanisms, remains underexplored. Furthermore, RM is susceptible to reward hacking in post-training. To mitigate these limitations, we propose SoliReward, a systematic framework for video RM training. Our framework first sources high-quality, cost-efficient data via single-item binary annotations, then constructs preference pairs using a cross-prompt pairing strategy. Architecturally, we employ a Hierarchical Progressive Query Attention mechanism to enhance feature aggregation. Finally, we introduce a modified BT loss that explicitly accommodates win-tie scenarios. This approach regularizes the RM's score distribution for positive samples, providing more nuanced preference signals to alleviate over-focus on a small number of top-scoring samples. Our approach is validated on benchmarks evaluating physical plausibility, subject deformity, and semantic alignment, demonstrating improvements in direct RM evaluation metrics and in the efficacy of post-training on video generation models. Code and benchmark will be publicly available.

Large Language Models (LLMs) have achieved remarkable progress in reasoning, yet sometimes produce responses that are suboptimal for users in tasks such as writing, information seeking, or providing practical guidance. Conventional alignment practices typically assume that maximizing model reward also maximizes user welfare, but this assumption frequently fails in practice: models may over-clarify or generate overly verbose reasoning when users prefer concise answers. Such behaviors resemble the prisoner's dilemma, where individually rational choices lead to socially suboptimal outcomes. The fundamental challenge is the lack of a principled decision making mechanism that mutually benefits both the LLM and the user. We propose Game-Theoretic Alignment (GTAlign), an alignment framework that integrates game-theoretic decision making into both reasoning and training. During reasoning, the model explicitly treats user-LLM interaction as a strategic game: it constructs payoff matrices within its reasoning chain to estimate welfare for both itself and the user, and then selects actions that are mutually beneficial. During training, we introduce a mutual welfare reward that reinforces cooperative responses, aligning model behavior with socially efficient outcomes. In addition, we introduce an inference technique that leverages game-theoretic reasoning to dynamically adapt LLM's response when pricing policies of LLM service change. Extensive experiments demonstrate that GTAlign substantially improves reasoning efficiency, answer quality, and mutual welfare compared to baselines across diverse tasks. The code is available at https://github.com/ulab-uiuc/GTAlign .

Recent research has made significant strides in applying alignment techniques to enhance the helpfulness and harmlessness of large language models (LLMs) in accordance with human intentions. In this paper, we argue for the importance of alignment for honesty, ensuring that LLMs proactively refuse to answer questions when they lack knowledge, while still not being overly conservative. However, a pivotal aspect of alignment for honesty involves discerning the limits of an LLM's knowledge, which is far from straightforward. This challenge demands comprehensive solutions in terms of metric development, benchmark creation, and training methodologies. In this paper, we address these challenges by first establishing a precise problem definition and defining ``honesty'' inspired by the Analects of Confucius. This serves as a cornerstone for developing metrics that effectively measure an LLM's honesty by quantifying its progress post-alignment. Furthermore, we introduce a flexible training framework which is further instantiated by several efficient fine-tuning techniques that emphasize honesty without sacrificing performance on other tasks. Our extensive experiments reveal that these aligned models show a marked increase in honesty, as indicated by our proposed metrics. We open-source a wealth of resources to facilitate future research at https://github.com/GAIR-NLP/alignment-for-honesty, including honesty-aligned models, training and evaluation datasets for honesty alignment, concept glossary, as well as all relevant source code.

The alignment tuning process of large language models (LLMs) typically involves instruction learning through supervised fine-tuning (SFT) and preference tuning via reinforcement learning from human feedback (RLHF). A recent study, LIMA (Zhou et al. 2023), shows that using merely 1K examples for SFT can achieve significant alignment performance as well, suggesting that the effect of alignment tuning might be "superficial." This raises questions about how exactly the alignment tuning transforms a base LLM. We analyze the effect of alignment tuning by examining the token distribution shift between base LLMs and their aligned counterpart. Our findings reveal that base LLMs and their alignment-tuned versions perform nearly identically in decoding on the majority of token positions. Most distribution shifts occur with stylistic tokens. These direct evidence strongly supports the Superficial Alignment Hypothesis suggested by LIMA. Based on these findings, we rethink the alignment of LLMs by posing the research question: how effectively can we align base LLMs without SFT or RLHF? To address this, we introduce a simple, tuning-free alignment method, URIAL. URIAL achieves effective alignment purely through in-context learning (ICL) with base LLMs, requiring as few as three constant stylistic examples and a system prompt. We conduct a fine-grained and interpretable evaluation on a diverse set of examples, named JUST-EVAL-INSTRUCT. Results demonstrate that base LLMs with URIAL can match or even surpass the performance of LLMs aligned with SFT or SFT+RLHF. We show that the gap between tuning-free and tuning-based alignment methods can be significantly reduced through strategic prompting and ICL. Our findings on the superficial nature of alignment tuning and results with URIAL suggest that deeper analysis and theoretical understanding of alignment is crucial to future LLM research.

Social deduction games like Werewolf combine language, reasoning, and strategy, providing a testbed for studying natural language and social intelligence. However, most studies reduce the game to LLM-based self-play, yielding templated utterances and anecdotal cases that overlook the richness of social gameplay. Evaluation further relies on coarse metrics such as survival time or subjective scoring due to the lack of quality reference data. To address these gaps, we curate a high-quality, human-verified multimodal Werewolf dataset containing over 100 hours of video, 32.4M utterance tokens, and 15 rule variants. Based on this dataset, we propose a novel strategy-alignment evaluation that leverages the winning faction's strategies as ground truth in two stages: 1) Speech evaluation, formulated as multiple-choice-style tasks that assess whether the model can adopt appropriate stances across five dimensions of social ability; and 2) Decision evaluation, which assesses the model's voting choices and opponent-role inferences. This framework enables a fine-grained evaluation of models' linguistic and reasoning capabilities, while capturing their ability to generate strategically coherent gameplay. Our experiments show that state-of-the-art LLMs show diverse performance, with roughly half remain below 0.50, revealing clear gaps in deception and counterfactual reasoning. We hope our dataset further inspires research on language, reasoning, and strategy in multi-agent interaction.

Recent studies have demonstrated significant progress in aligning text-to-image diffusion models with human preference via Reinforcement Learning from Human Feedback. However, while existing methods achieve high scores on automated reward metrics, they often lead to Preference Mode Collapse (PMC)-a specific form of reward hacking where models converge on narrow, high-scoring outputs (e.g., images with monolithic styles or pervasive overexposure), severely degrading generative diversity. In this work, we introduce and quantify this phenomenon, proposing DivGenBench, a novel benchmark designed to measure the extent of PMC. We posit that this collapse is driven by over-optimization along the reward model's inherent biases. Building on this analysis, we propose Directional Decoupling Alignment (D^2-Align), a novel framework that mitigates PMC by directionally correcting the reward signal. Specifically, our method first learns a directional correction within the reward model's embedding space while keeping the model frozen. This correction is then applied to the reward signal during the optimization process, preventing the model from collapsing into specific modes and thereby maintaining diversity. Our comprehensive evaluation, combining qualitative analysis with quantitative metrics for both quality and diversity, reveals that D^2-Align achieves superior alignment with human preference.

Aligning Large Language Models (LLMs) to human preferences through preference optimization has been crucial but labor-intensive, necessitating for each prompt a comparison of both a chosen and a rejected text completion by evaluators. Recently, Kahneman-Tversky Optimization (KTO) has demonstrated that LLMs can be aligned using merely binary "thumbs-up" or "thumbs-down" signals on each prompt-completion pair. In this paper, we present theoretical foundations to explain the successful alignment achieved through these binary signals. Our analysis uncovers a new perspective: optimizing a binary classifier, whose logit is a reward, implicitly induces minimizing the Direct Preference Optimization (DPO) loss. In the process of this discovery, we identified two techniques for effective alignment: reward shift and underlying distribution matching. Consequently, we propose a new algorithm, Binary Classifier Optimization, that integrates the techniques. We validate our methodology in two settings: first, on a paired preference dataset, where our method performs on par with DPO and KTO; and second, on binary signal datasets simulating real-world conditions with divergent underlying distributions between thumbs-up and thumbs-down data. Our model consistently demonstrates effective and robust alignment across two base LLMs and three different binary signal datasets, showcasing the strength of our approach to learning from binary feedback.

Fine-tuning text-to-image models with reward functions trained on human feedback data has proven effective for aligning model behavior with human intent. However, excessive optimization with such reward models, which serve as mere proxy objectives, can compromise the performance of fine-tuned models, a phenomenon known as reward overoptimization. To investigate this issue in depth, we introduce the Text-Image Alignment Assessment (TIA2) benchmark, which comprises a diverse collection of text prompts, images, and human annotations. Our evaluation of several state-of-the-art reward models on this benchmark reveals their frequent misalignment with human assessment. We empirically demonstrate that overoptimization occurs notably when a poorly aligned reward model is used as the fine-tuning objective. To address this, we propose TextNorm, a simple method that enhances alignment based on a measure of reward model confidence estimated across a set of semantically contrastive text prompts. We demonstrate that incorporating the confidence-calibrated rewards in fine-tuning effectively reduces overoptimization, resulting in twice as many wins in human evaluation for text-image alignment compared against the baseline reward models.

Loss explosions in training deep neural networks can nullify multi-million dollar training runs. Conventional monitoring metrics like weight and gradient norms are often lagging and ambiguous predictors, as their values vary dramatically across different models and even between layers of the same model, making it difficult to establish a unified standard for detecting impending failure. We introduce Spectral Alignment (SA), a novel, theoretically-grounded metric that monitors the distributional alignment between layer inputs and the principal singular vectors of weight matrices. We show that a collapse in the sign diversity of this alignment is a powerful early predictor of representational collapse and training divergence. Empirical results on language models demonstrate that monitoring the SA distribution provides a significantly earlier and clearer warning of loss explosions than traditional scalar metrics. SA's low computational overhead makes it a practical tool for safeguarding model training.

A simple and effective method for the alignment of generative models is the best-of-n policy, where n samples are drawn from a base policy, and ranked based on a reward function, and the highest ranking one is selected. A commonly used analytical expression in the literature claims that the KL divergence between the best-of-n policy and the base policy is equal to log (n) - (n-1)/n. We disprove the validity of this claim, and show that it is an upper bound on the actual KL divergence. We also explore the tightness of this upper bound in different regimes. Finally, we propose a new estimator for the KL divergence and empirically show that it provides a tight approximation through a few examples.