AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

ACL 2026

GoViG: Goal-Conditioned Visual Navigation Instruction Generation via Multimodal Reasoning

Fengyi Wu, Yifei Dong, Yilong Dai, Guangyu Chen, Qifeng Wu, Huiting Huang, Hang Wang, Qi Dai, Alexander G Hauptmann, Zhi-Qi Cheng

We introduce Goal-Conditioned Visual Navigation Instruction Generation (GoViG), a new task that aims to generate contextually coherent navigation instructions solely from egocentric visual observations of initial and goal states. Unlike prior work relying on structured inputs, such as semantic annotations or environmental maps, GoViG exclusively leverages raw egocentric visual data, improving adaptability to unseen and unstructured environments. Our method addresses this task by decomposing it into two interconnected subtasks: (1) navigation visualization, predicting intermediate visual states bridging the initial and goal views; and (2) instruction generation, synthesizing coherent instructions grounded in observed and anticipated visuals. Both subtasks are integrated within an autoregressive multimodal LLM trained with tailored objectives to ensure spatial accuracy and linguistic clarity. Furthermore, we introduce two multimodal reasoning strategies, one-pass and interleaved reasoning, to mimic incremental human navigation cognition. To comprehensively evaluate our method, we propose the R2R-Goal dataset, combining diverse synthetic and real-world trajectories. Empirical results demonstrate significant performance improvements over state-of-the-art methods in BLEU-4 and CIDEr scores along with robust cross-domain generalization. Our project is available at https://github.com/F1y1113/GoViG.

ACL 2026

Invocation Refiner: A Plug-and-Play Module for Rectifying LLM Tool Invocations

Qirui Jiao, Dian Jiao, Nan Du, Ying Shen, Liang Lin

Large Language Models (LLMs) have shown remarkable capabilities in Tool-Integrated Reasoning (TIR). However, the practical application is often hindered by frequent errors in tool invocations, such as incorrect parameters or malformed formats. Prevailing training paradigms, such as Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL), can mitigate these issues but require modification on the base LLM. This lack of modularity necessitates extensive retraining when deploying the system across different base models. To address the limitation, we introduce the Invocation Refiner, a specialized post-processing module designed to enhance the tool-use reliability of base LLMs without directly training on them. The Refiner takes the output from a frozen upstream LLM and the user’s query as input, performing independent reasoning to rectify the invocation. We construct a dedicated training dataset and train this module using an advanced RL algorithm. On a diverse set of tool-use and reasoning benchmarks, our Refiner improves task completion rates and invocation accuracy over the raw outputs of various upstream LLMs. This highlights our Refiner as a plug-and-play solution for improving the operational reliability of LLM-based agents. We release our code to facilitate future research.

ACL 2026

FreeChunker: A Cross-Granularity Chunking Framework

Wenxuan Zhang, Yuan-Hao Jiang, Yang Cao, Yonghe Wu

Chunking strategies significantly impact the effectiveness of Retrieval-Augmented Generation (RAG) systems. Existing methods operate within fixed-granularity paradigms that rely on static boundary identification, limiting their adaptability to diverse query requirements. This paper presents FreeChunker, a Cross-Granularity Encoding Framework that fundamentally transforms the traditional chunking paradigm: the framework treats sentences as atomic units and shifts from static chunk segmentation to flexible retrieval supporting arbitrary sentence combinations. This paradigm shift not only significantly avoids the computational overhead required for semantic boundary detection, but also enhances adaptability to complex queries. Experimental evaluation on LongBench V2 demonstrates that FreeChunker possesses significant advantages in both retrieval performance and time efficiency compared to existing chunking methods. The pre-trained models and codes are available at https://github.com/mazehart/FreeChunker.

ACL 2026

Spatiotemporal Sycophancy: Negation-Based Gaslighting in Video Large Language Models

Ziyao Tang, Pengkun Jiao, Bin Zhu, Huiyan Qi, Jingjing Chen, Yu-Gang Jiang

Video Large Language Models (Vid-LLMs) have demonstrated remarkable performance in video understanding tasks, yet their robustness under conversational interaction remains largely underexplored. In this paper, we identify spatiotemporal sycophancy, a failure mode in which Vid-LLMs retract initially correct, visually grounded judgments and conform to misleading user feedback under negation-based gaslighting. Rather than merely changing their answers, the models often fabricate unsupported temporal or spatial explanations to justify incorrect revisions. To systematically investigate this phenomenon, we propose a negation-based gaslighting evaluation framework and introduce GasVideo-1000, a curated benchmark designed to probe spatiotemporal sycophancy with clear visual grounding and temporal reasoning requirements. We evaluate a broad range of state-of-the-art open-source and proprietary Vid-LLMs across diverse video understanding tasks. Extensive experiments reveal that vulnerability to negation-based gaslighting is pervasive and severe, even among models with strong baseline performance. While prompt-level grounding constraints can partially mitigate this behavior, they do not reliably prevent hallucinated justifications or belief reversal. Our results indicate that current Vid-LLMs lack robust mechanisms for maintaining grounded spatiotemporal beliefs under adversarial conversational feedback.

ACL 2026

AlphaEdit+: Model Editing in the Presence of Conflicting and Inconsistent Knowledge

Qing Liu, Jianhao Zhang, Ou Wu, Michael Ng, Yi Du

Knowledge editing is a crucial technique for daily updates in LLMs, requiring a balance between accurately modifying incorrect knowledge and preserving existing information. The recently proposed AlphaEdit method achieves competitive editing performance by updating parameters under null-space constraints. However, our theoretical analysis reveals that AlphaEdit struggles with high knowledge conflicts and inconsistencies during editing. To address this, we propose a new editing method AlphaEdit+, featuring three key improvements: 1) relaxing null-space constraints by adding a matrix perturbation through optimization to resolve conflicts between new and preserved knowledge; 2) introducing a weighting scheme on previously updated knowledge constraints to mitigate conflicts between new and historical editing; 3) developing a value smoothing algorithm to resolve high knowledge inconsistencies. These enhancements collectively ensure robust editing while maintaining model coherence. Comprehensive experiments show that our approach AlphaEdit+ not only resolves the brittleness of the original method on carefully constructed challenging datasets but also outperforms AlphaEdit on existing benchmark datasets.

ACL 2026

AgenticEval: Toward Agentic and Self-Evolving Safety Evaluation of Large Language Models

Yixu Wang, Xin Wang, Yang Yao, Xinyuan Li, Xibang Yang, Yan Teng, Xingjun Ma, Yingchun Wang

The rapid integration of Large Language Models (LLMs) into high-stakes domains necessitates reliable safety and compliance evaluation. However, existing static benchmarks are ill-equipped to address the dynamic nature of AI risks and evolving regulations, creating a critical safety gap. This paper introduces a new paradigm of agentic safety evaluation, reframing evaluation as a continuous and self-evolving process rather than a one-time audit. We then propose a novel multi-agent framework AgenticEval, which autonomously ingests unstructured policy documents to generate and perpetually evolve a comprehensive safety benchmark. AgenticEval leverages a synergistic pipeline of specialized agents and incorporates a Self-evolving Evaluation loop, where the system learns from evaluation results to craft progressively more sophisticated and targeted test cases. Our experiments demonstrate the effectiveness of AgenticEval, showing a consistent decline in model safety as the evaluation hardens. For instance, GPT-5’s safety rate on the EU AI Act drops from 72.50% to 36.36% over successive iterations. These findings reveal the limitations of static assessments and highlight our framework’s ability to uncover deep vulnerabilities missed by traditional methods, underscoring the urgent need for dynamic evaluation ecosystems to ensure the safe and responsible deployment of advanced AI.

ACL 2026

HSUGA: LLM-Enhanced Recommendation with Hierarchical Semantic Understanding and Group-Aware Alignment

Guorui Li, Dugang Liu, Lei Li, Xing Tang, Zhong Ming

Large language model (LLM)-enhanced sequential recommendation typically aims to improve two core components: user semantic embedding extraction and utilization. Despite promising results, existing methods still have two limitations: 1) In the extraction stage, most methods directly input long interaction sequence fragments into LLM for preference summarization. However, excessively long sequences increase inference difficulty, making it challenging to infer accurate user embeddings reliably. 2) In the utilization stage, most methods employ the same semantic embedding utilization strategy for all users, neglecting the differences caused by user activity levels, leading to suboptimal performance. To address these issues, we propose HSUGA, which introduces a simple yet effective plugin for each of the two core components: Hierarchical Semantic Understanding (HSU) and Group-Aware Alignment (GAA). HSU performs a staged two-phase preference mining and models preference evolution through constrained editing operations, thereby improving the reliability of user semantic extraction. GAA adjusts the semantic utilization intensity based on user activity levels, providing weaker alignment for active users and stronger guidance for users with sparse historical data. Finally, extensive experiments on three benchmark datasets demonstrate the effectiveness and compatibility of HSUGA.

ACL 2026

Meeseeks: A Feedback-Driven, Iterative Self-Correction Benchmark evaluating LLMs’ Instruction Following Capability

Jiaming Wang, Yunke Zhao, Peng Ding, Jun Kuang, Yibin Shen, Zhe Tang, Yilin Jin, Zongyu Wang, Xiaoyu Li, Xuezhi Cao

The capability to precisely adhere to instructions is a cornerstone for Large Language Models (LLMs) to function as dependable agents in real-world scenarios. However, confronted with complex prompts, LLMs frequently encounter difficulties in fulfilling all specified requirements within a single response. Drawing inspiration from recent advancements in Chain-of-Thought (CoT) prompting and self-correction methodologies, we introduce Meeseeks, a fully automated iterative instruction-following benchmark equipped with an integrated feedback mechanism. Meeseeks identifies erroneous components in model responses and provides corresponding feedback accurately, thereby iteratively guiding the model toward self-correction. The dataset contains over 700 curated instances annotated by 32 distinct capability tags in Chinese and English. Extensive experimental results reveal that different state-of-the-art commercial and open-source LLMs exhibit vastly disparate performance, and even after 20 turns of iterative feedback-driven self-correction, nearly all models demonstrate suboptimal performance. We conducted comprehensive analysis and uncovered numerous common issues prevalent in current state-of-the-art models, as well as several counterintuitive phenomena. Meeseeks has been open-sourced on https://github.com/ADoublLEN/Meeseeks.

ACL 2026

How to Allocate, How to Learn? Dynamic Rollout Allocation and Advantage Modulation for Policy Optimization

Yangyi Fang, Jiaye Lin, Xiaoliang Fu, Cong Qin, Haolin Shi, Chaowen Hu, Lu Pan, Ke Zeng, Xunliang Cai

Reinforcement Learning with Verifiable Rewards (RLVR) has proven effective for Large Language Model (LLM) reasoning, yet current methods face key challenges in resource allocation and policy optimization dynamics: (i) uniform rollout allocation ignores gradient variance heterogeneity across problems, and (ii) the softmax policy structure causes gradient attenuation for high-confidence correct actions, while excessive gradient updates may destabilize training. Therefore, we propose DynaMO, a theoretically-grounded dual-pronged optimization framework. At the sequence level, we prove that uniform allocation is suboptimal and derive variance-minimizing allocation from the first principle, establishing Bernoulli variance as a computable proxy for gradient informativeness. At the token level, we develop gradient-aware advantage modulation grounded in theoretical analysis of gradient magnitude bounds. Our framework compensates for gradient attenuation of high-confidence correct actions while utilizing entropy changes as computable indicators to stabilize excessive update magnitudes. Extensive experiments conducted on a diverse range of mathematical reasoning benchmarks demonstrate consistent improvements over strong RLVR baselines. Our implementation is available at: https://github.com/FlyTune/DynaMO-RL.

ACL 2026

Learning to Translate by Translating: Stabilizing the Dual Loop via Semantic-Aware Self-Evolution

Kui Liu, Mingming Yin, Zuoli Tang, Zihao Li, Chilin Fu, Xiaolu Zhang, JUN ZHOU, Lixin Zou, Chenliang Li

Despite the remarkable success of Large Language Models (LLMs) in Machine Translation (MT), the scarcity of high-quality parallel corpora and the prohibitive cost of their acquisition constrain scalability. To this end, we propose Learning to Translate by Translating (LTT), an LLM-driven dual-learning framework that enables autonomous translation, achieving an 80.42% performance improvement over the base model. By adapting the cycle-consistency principle to the generative paradigm, LTT eliminates the need for parallel data. It employs a robust semantic-aware reward function that balances adequacy with reconstruction fidelity, effectively mitigating the reward hacking issues inherent in traditional unsupervised MT. Relying solely on monolingual data, our 8B model consistently outperforms significantly larger models (70B+) in low-resource settings and achieves parity with state-of-the-art supervised baselines on mainstream benchmarks. LTT thus offers a scalable, data-efficient paradigm for autonomous machine translation.

ACL 2026

SAMem: State-Aware Memory as a Fine-Grained Memory for LLM Agents in Decision-Making

Tong Wang, Pei Xu, Shiyue Cao, Likun Yang, Daipeng Li, Jianbin Jiao, Kaiqi Huang

Existing LLM-based agents primarily utilize coarse-grained experiential memory, where experiences are retrieved based on global task or scene context. While effective in simple settings, such coarse-grained memory lacks the situational alignment required for complex multi-step decision-making. As a result, recalled experiences often fail to match the agent’s current state, blurring reasoning focus and leading to inaccurate decisions at critical steps. To this end, we propose State-Aware memory(SAMem), a new fine-grained memory paradigm for LLM agents that explicitly aligns memory retrieval with the current state. Instead of storing and reusing globally shared experiences, SAMem organizes memory at the level of state-specific reasoning thoughts, enabling the agent to retrieve only the most relevant experience for the current decision context. This state-conditioned memory allows the agent to focus on the most informative reasoning cues at each step, rather than being distracted by task-level but state-misaligned guidance. Extensive experiments on complex decision-making benchmarks demonstrate that SAMem outperforms existing experiential memory approaches, achieving superior performance and substantially improved task-solving efficiency. These results indicate that state-aware, fine-grained memory enhances the decision-making capabilities of LLM agents.

ACL 2026

MultiCodeAttack: Iterative Jailbreak Attacking on LLMs with Multi-Code Prompt Injection

Weifeng Sun, Meng Yan, Zhou Yang, Yuchen Chen, Song Sun, David Lo

Large Language Models (LLMs) demonstrate strong generalization capabilities but remain vulnerable to jailbreak attacks that induce restricted text or malicious code generation.Recent structured jailbreaks embed adversarial intent into code-like templates and have demonstrated promising effectiveness.However, existing approaches typically operate within a fixed template design and a single programming language, without considering language diversity or adaptive template evolution, thereby limiting the exploration of cross-language jailbreak behaviors.In this paper, we present MultiCodeAttack, a structured jailbreak framework that systematically explores and optimizes multi-language code templates.MultiCodeAttack maintains a diverse template library across programming languages, dynamically selects languages with higher attack effectiveness via a multi-armed bandit strategy, and evolves templates through semantic-preserving mutation guided by response-aware signals.Extensive experiments on 8 LLMs show that MultiCodeAttack outperforms existing jailbreak baselines, achieving 28.23%–832.59% higher harmful text generation.On malicious code generation across 11 LLMs, MultiCodeAttack produces up to 136.22% more malicious outputs than the baseline methods.Our code is available at https://anonymous.4open.science/r/MultiCodeAttack/.

ACL 2026

Large-Scale Multimodal Knowledge Graph about Classical Chinese Poetry: Fine-grained Method and Comprehensive Evaluation

Shuo Wang, Qing Zhu, Yang Xiao, Minglong Lei

Classical Chinese poetry is a treasured cultural heritage of humanity, attracting extensive research interest. However, the study of classical Chinese poetry is hindered by the lack of open, large-scale, and fine-grained multimodal datasets.Prior datasets are either limited by modality constraints, dataset size, or the level of dataset refinement, making them inadequate for effectively supporting studies and the development of applications in classical Chinese poetry.To address these issues, we propose a method for constructing a large-scale and fine-grained multimodal knowledge graph of classical Chinese poetry. We first design an informative ontology graph for classical Chinese poetry and comprehensively collect knowledge about poetry based on it. Furthermore, the method leverages knowledge augmentation, prompt optimization, and text-image alignment to acquire comprehensive, fine-grained knowledge. Both qualitative and quantitative evaluations are conducted on the Multimodal Knowledge Graph of Classical Chinese Poetry (CPMK), highlighting its comprehensiveness and high quality.We also conduct downstream evaluations on four tasks: poetry question answering, poetry theme classification, poetry-image retrieval, and rigid-formats poetry generation.Significant results are achieved across all four tasks, demonstrating CPMK’s effectiveness in supporting research on Chinese poetry.CPMK will be released to promote research in Chinese culture.

ACL 2026

Mistake Notebook Learning: Batch-Clustered Failures for Training-Free Agent Adaptation

Xuanbo Su, Yingfang Zhang, Hao Luo, Xiaoteng Liu, Leo Huang

With the growing adoption of Large Language Model (LLM) agents in persistent, real-world roles, they naturally encounter continuous streams of tasks and inevitable failures. A key limitation, however, is their inability to systematically learn from these mistakes, forcing them to repeat identical errors in similar contexts. Unlike prior training-free methods that primarily store raw instance-level experience or focus on retrieving successful trajectories, we propose Mistake Notebook Learning (MNL), a novel memory framework that enables agents to self-curate generalizable guidance from batch-clustered failures. This mechanism allows agents to distill shared error patterns into structured “mistake notes”, updating an external memory only when batch performance improves to ensure stability. To further amplify adaptability, we integrate MNL with test-time scaling, leveraging aggregated failure patterns to actively steer the search process away from known pitfalls. Experiments on mathematical reasoning, Text-to-SQL, and interactive agent benchmarks show that MNL achieves competitive performance compared to existing memory mechanisms in both effectiveness and efficiency. These findings position structured mistake abstraction as a critical lever for robust agent evolution, enabling continuous improvement without the cost of parameter updates.

ACL 2026

LPO: Towards Accurate GUI Agent Interaction via Location Preference Optimization

Jiaqi Tang, Yu Xia, Yi-Feng Wu, Yuwei Hu, Chen Yuhui, Qing-Guo Chen, Xiaogang Xu, Xiangyu Wu, Hao Lu, YanQing Ma, Shiyin Lu, Qifeng Chen

The advent of autonomous agents is transforming interactions with Graphical User Interfaces (GUIs) by employing natural language as a powerful intermediary. Despite the predominance of supervised fine-tuning (SFT) methods in current GUI agents for achieving spatial localization, these methods face substantial challenges due to their limited capacity to accurately perceive positional data. Existing strategies, such as reinforcement learning, often fail to assess positional accuracy effectively, thereby restricting their utility. In response, we introduce Location Preference Optimization (LPO), a novel approach that leverages locational data to optimize interaction preferences. LPO uses information entropy to predict interaction positions by focusing on zones rich in information. Besides, we further introduce a dynamic location reward function based on physical distance, reflecting the varying importance of interaction positions. Supported by Group Relative Preference Optimization (GRPO), LPO facilitates an extensive exploration of GUI environments and significantly enhances interaction precision. Comprehensive experiments demonstrate LPO’s superior performance, achieving SOTA results across both offline benchmarks and real-world online evaluations.

ACL 2026

Representation-Guided Parameter-Efficient LLM Unlearning

Zeguan Xiao, Lang Mo, Yun Chen, Lei Yang, Jiehui Zhao, Lili Yang, Guanhua Chen

Large Language Models (LLMs) often memorize sensitive or harmful information, necessitating effective machine unlearning techniques. While existing parameter-efficient unlearning methods have shown promise, they still struggle with the forget-retain trade-off. This can be attributed to their reliance on parameter importance metrics to identify parameters that are important exclusively for forget set, which is fundamentally limited by the superposition phenomenon. Due to the polysemantic nature of LLMs parameters, such an importance metric may struggle to disentangle parameters associated with forget and retain sets. In this work, we propose Representation-Guided Low-rank Unlearning (ReGLU), a novel approach that leverages the geometric properties of representation spaces to achieve robust and precise unlearning. First, we develop a representation-guided initialization for LoRA that identifies the optimal subspace for selective forgetting. Second, we introduce a regularization loss that constrains the outputs of the LoRA update to lie in the orthogonal complement of the retain set’s representation subspace, thereby minimizing interference with the model’s performance on the retain set. We evaluate ReGLU on the TOFU and WMDP benchmarks across multiple models. Our results demonstrate that ReGLU consistently outperforms state-of-the-art baselines, achieving superior unlearning quality while maintaining higher model utility.

ACL 2026

MalURLBench: A Benchmark Evaluating Agents’ Vulnerabilities When Processing Web URLs

Dezhang Kong, Zhuxi Wu, Shiqi Liu, ZhiCheng Tan, Kuichen Lu, Minghao Li, Qichen Liu, Shengyu Chu, Zhenhua Xu, Xuan Liu, Meng Han

LLM-based web agents have become increasingly popular for their utility in daily life and work. However, they exhibit critical vulnerabilities when processing malicious URLs: accepting a disguised malicious URL enables subsequent access to unsafe webpages, which can cause severe damage to service providers and users. Despite this risk, no benchmark currently targets this emerging threat. To address this gap, we propose MalURLBench, the first benchmark for evaluating LLMs’ vulnerabilities to malicious URLs. MalURLBench contains 61,845 attack instances spanning 10 real-world scenarios and 7 categories of real malicious websites. Experiments with 12 popular LLMs reveal that existing models struggle to detect elaborately disguised malicious URLs. We further identify and analyze key factors that impact attack success rates and propose URLGuard, a lightweight defense module. We believe this work will provide a foundational resource for advancing the security of web agents.

ACL 2026

SOP-Maze: Evaluating Large Language Models on Complicated Business Standard Operating Procedures

Jiaming Wang, Zhe Tang, Zehao Jin, Hefei Chen, Yilin Jin, Peng Ding, Xiaoyu Li, Xuezhi Cao

As large language models (LLMs) are widely deployed as domain-specific agents, many benchmarks have been proposed to evaluate their ability to follow instructions and make decisions in real-world scenarios. However, business scenarios often involve complex standard operating procedures (SOPs), and the evaluation of LLM capabilities in such contexts has not been fully explored. To bridge this gap, we propose SOP-Maze, a benchmark constructed from real-world business data and adapted into a collection of 397 instances and 3422 subtasks from 23 complex SOP scenarios. We further categorize SOP tasks into two broad classes: Lateral Root System (LRS), representing wide-option tasks that demand precise selection; and Heart Root System (HRS), which emphasizes deep logical reasoning with complex branches. Extensive experiments reveal that nearly all state-of-the-art models struggle with SOP-Maze. We conduct a comprehensive analysis and identify three key error categories: (i) route blindness: difficulty following procedures; (ii) conversational fragility: inability to handle real dialogue nuances; and (iii) calculation errors: mistakes in time or arithmetic reasoning under complex contexts. The systematic study explores LLM performance across SOP tasks that challenge both breadth and depth, offering new insights for improving model capabilities. We have open-sourced our work on the anonymous link: https://github.com/meituan-longcat/SOP-Maze.

ACL 2026

FABLE: Fine-grained Fact Anchoring for Unstructured Model Editing

Peng Wang, biyu zhou, Xuehai Tang, Jizhong Han, Songlin Hu

Unstructured model editing aims to update models with real-world text, yet existing methods often memorize text holistically without reliable fine-grained fact access. To address this, we propose FABLE, a hierarchical framework that decouples fine-grained fact injection from holistic text generation. FABLE follows a two-stage, fact-first strategy: discrete facts are anchored in shallow layers, followed by minimal updates to deeper layers to produce coherent text. This decoupling resolves the mismatch between holistic recall and fine-grained fact access, reflecting the unidirectional Transformer flow in which surface-form generation amplifies rather than corrects underlying fact representations. We also introduce UnFine, a diagnostic benchmark with fine-grained question–answer pairs and fact-level metrics for systematic evaluation. Experiments show that FABLE substantially improves fine-grained question answering while maintaining state-of-the-art holistic editing performance. Our code is publicly available at https://anonymous.4open.science/r/FABLE-B59E.

ACL 2026

Towards Bridging the Reward-Generation Gap in Direct Alignment Algorithms

Zeguan Xiao, Yun Chen, Jian Yang, Guanhua Chen, Ke Tang

Direct Alignment Algorithms (DAAs), such as Direct Preference Optimization (DPO) and Simple Preference Optimization (SimPO), have emerged as efficient alternatives to Reinforcement Learning from Human Feedback (RLHF) algorithms for aligning large language models (LLMs) with human preferences. However, DAAs suffer from a fundamental limitation we identify as the “reward-generation gap”—a discrepancy between training objectives and autoregressive decoding dynamics. In this paper, we consider that one contributor to the reward-generation gap is the mismatch between the inherent importance of prefix tokens during the LLM generation process and how this importance is reflected in the implicit reward functions of DAAs. To bridge the gap, we adopt a token-level MDP perspective of DAAs to analyze its limitations and introduce a simple yet effective approach called Prefix-Oriented Equal-length Training (POET), which truncates both preferred and dispreferred responses to match the shorter one’s length. We conduct experiments with DPO and SimPO, two representative DAAs, demonstrating that POET improves over their standard implementations, achieving up to 11.8 points in AlpacaEval 2 and overall improvements across downstream tasks. These results underscore the need to mitigate the reward-generation gap in DAAs by better aligning training objectives with autoregressive decoding dynamics.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.