Recently, large-scale pre-trained speech encoders and Large Language Models (LLMs) have been released, which show state-of-the-art performance on a range of spoken language processing tasks, including Automatic Speech Recognition (ASR). To effectively combine both models for better performance, continuous speech prompts and ASR error correction have been adopted. However, these methods are prone to suboptimal performance or are inflexible. In this paper, we propose a new paradigm, LegoSLM, that bridges speech encoders and LLMs using the ASR posterior matrices. The speech encoder is trained to generate Connectionist Temporal Classification (CTC) posteriors over the LLM vocabulary, which are used to reconstruct pseudo-audio embeddings by computing a weighted sum of the LLM input embeddings. These embeddings are concatenated with text embeddings in the LLM input space. Using the well-performing USM and Gemma models as an example, we demonstrate that our proposed LegoSLM method yields good performance on both ASR and speech translation tasks. By connecting USM with Gemma models, we can get an average of 49% WER reduction (WERR) over the USM-CTC baseline on 8 MLS testsets. The trained model also exhibits modularity in a range of settings – after fine-tuning the Gemma model weights, the speech encoder can be switched and combined with the LLM in a zero-shot fashion. Additionally, we propose to control the decode-time influence of the USM and LLM using a softmax temperature, which shows effectiveness in domain adaptation.
论文检索
输入标题、作者或关键词，从 7,591 篇学术成果中精准定位
No Free Lunch: Retrieval-Augmented Generation Undermines Fairness in LLMs, Even for Vigilant Users
PDF ↗Retrieval-Augmented Generation (RAG) is widely adopted for its effectiveness and cost-efficiency in mitigating hallucinations and enhancing the domain-specific generation capabilities of large language models (LLMs). However, is this effectiveness and cost-efficiency truly a free lunch? In this study, we comprehensively investigate the fairness costs associated with RAG by proposing a practical three-level threat model from the perspective of user awareness of fairness. Specifically, varying levels of user fairness awareness result in different degrees of fairness censorship on external datasets. We examine the fairness implications of RAG using uncensored, partially censored, and fully censored datasets. Our experiments demonstrate that fairness alignment can be easily undermined through RAG without the need for fine-tuning or retraining. Even with fully censored and supposedly unbiased external datasets, RAG would still lead to biased outputs. Our findings underscore the limitations of current alignment methods in the context of RAG-based LLMs and highlight the urgent need for new strategies to ensure fairness. We propose potential mitigations and call for further research to develop robust fairness safeguards in RAG-based LLMs.
Long-form outline generation for expository articles requires both comprehensive knowledge coverage and logical coherence, which is essential for creating detailed Wikipedia-like content. However, existing methods face critical limitations: outlines generated in the pre-writing stage often have low knowledge density and lack detail, while retrieval-augmented approaches struggle to maintain logical coherence across retrieved information. Additionally, unlike human writers who can iteratively improve through peer feedback and reference similar topics, current approaches lack effective mechanisms for systematic outline refinement. To address these challenges, we propose Logic, a Long-form Outline Generation system via Imitative and Critical self-refinement that mimics human writers’ refinement process. Logic establishes a coherent planning framework and structured knowledge base, learns from similar topic outlines through imitation, and continuously improves through model-based critique. Experiments on FreshWiki and our dataset WikiOutline show that, compared to the best baseline, Logic’s long-form outlines are more organized (with increases of 22.85% and 21.65% respectively) and more logically coherent (with increases of 16.19% and 12.24% respectively). Human evaluation further validates Logic’s effectiveness in generating comprehensive and well-structured long-form outlines.
Assess and Prompt: A Generative RL Framework for Improving Engagement in Online Mental Health Communities
PDF ↗Online Mental Health Communities (OMHCs) provide crucial peer and expert support, yet many posts remain unanswered due to missing support attributes that signal the need for help. We present a novel framework that identifies these gaps and prompts users to enrich their posts, thereby improving engagement. To support this, we introduce REDDME, a new dataset of 4,760 posts from mental health subreddits annotated for the span and intensity of three key support attributes: event what happened?, effect what did the user experience?, and requirement what support they need?. Next, we devise a hierarchical taxonomy, CueTaxo, of support attributes for controlled question generation. Further, we propose MH-COPILOT, a reinforcement learning-based system that integrates (a) contextual attribute-span identification, (b) support attribute intensity classification, (c) controlled question generation via a hierarchical taxonomy, and (d) a verifier for reward modeling. Our model dynamically assesses posts for the presence/absence of support attributes, and generates targeted prompts to elicit missing information. Empirical results across four notable language models demonstrate significant improvements in attribute elicitation and user engagement. A human evaluation further validates the model’s effectiveness in real-world OMHC settings.
FOSSIL: Harnessing Feedback on Suboptimal Samples for Data-Efficient Generalisation with Imitation Learning for Embodied Vision-and-Language Tasks
PDF ↗Current approaches to embodied AI tend to learn policies from expert demonstrations. However, without a mechanism to evaluate the quality of demonstrated actions, they are limited to learning from optimal behaviour or risk replicating errors and inefficiencies. While reinforcement learning offers one alternative, the associated exploration typically results in sacrificing data efficiency. This work explores how agents trained with imitation learning can learn robust representations from both optimal and suboptimal demonstrations when given access to constructive language feedback as a means to contextualise different modes of behaviour. We directly provide language feedback embeddings as part of the input sequence into a Transformer-based policy, and optionally complement the traditional next action prediction objective with auxiliary self-supervised learning objectives for feedback prediction. We test our approach on a range of embodied Vision-and-Language tasks in our custom BabyAI-XGen environment and show significant improvements in agents’ compositional generalisation abilities and robustness, suggesting that our data-efficient method allows models to successfully convert suboptimal behaviour into learning opportunities. Overall, our results suggest that language feedback is a competitive and intuitive alternative to intermediate scalar rewards for language-specified embodied tasks.
Automatic content moderation is crucial to ensuring safety in social media. Language Model-based classifiers are increasingly adopted for this task, but it has been shown that they perpetuate racial and social biases. Even if several resources and benchmark corpora have been developed to challenge this issue, measuring the fairness of models in content moderation remains an open issue. In this work, we present an unsupervised approach that benchmarks models on the basis of their uncertainty in classifying messages annotated by people belonging to vulnerable groups. We use uncertainty, computed by means of the conformal prediction technique, as a proxy to analyze the bias of 11 models (LMs and LLMs) against women and non-white annotators and observe to what extent it diverges from metrics based on performance, such as the F1 score. The results show that some pre-trained models predict with high accuracy the labels coming from minority groups, even if the confidence in their prediction is low. Therefore, by measuring the confidence of models, we are able to see which groups of annotators are better represented in pre-trained models and lead the debiasing process of these models before their effective use.
Large language models (LLMs) have advanced the automation of data science workflows. Yet it remains unclear whether they can critically leverage external domain knowledge as human data scientists do in practice. To answer this question, we introduce AssistedDS (Assisted Data Science), a benchmark designed to systematically evaluate how LLMs handle domain knowledge in tabular prediction tasks. AssistedDS features both synthetic datasets with explicitly known generative mechanisms and real-world Kaggle competitions, each accompanied by curated bundles of helpful and adversarial documents. These documents provide domain-specific insights into data cleaning, feature engineering, and model selection. We assess state-of-the-art LLMs on their ability to discern and apply beneficial versus harmful domain knowledge, evaluating submission validity, information recall, and predictive performance. Our results demonstrate three key findings: (1) LLMs frequently exhibit an uncritical adoption of provided information, significantly impairing their predictive performance when adversarial content is introduced, (2) helpful guidance is often insufficient to counteract the negative influence of adversarial information, and (3) in Kaggle datasets, LLMs often make errors in handling time-series data, applying consistent feature engineering across different folds, and interpreting categorical variables correctly. These findings highlight a substantial gap in current models’ ability to critically evaluate and leverage expert knowledge, underscoring an essential research direction for developing more robust, knowledge-aware automated data science systems. Our data and code are publicly available [here](https://github.com/jeremyxianx/Assisted-DS).
Biology-Instructions: A Dataset and Benchmark for Multi-Omics Sequence Understanding Capability of Large Language Models
PDF ↗Large language models (LLMs) have shown remarkable capabilities in general domains, but their application to multi-omics biology remains underexplored. To address this gap, we introduce Biology-Instructions, the first large-scale instruction-tuning dataset for multi-omics biological sequences, including DNA, RNA, proteins, and multi-molecules. This dataset bridges LLMs and complex biological sequence-related tasks, enhancing their versatility and reasoning while maintaining conversational fluency. We also highlight significant limitations of current state-of-the-art LLMs on multi-omics tasks without specialized training. To overcome this, we propose ChatMultiOmics, a strong baseline with a novel three-stage training pipeline, demonstrating superior biological understanding through Biology-Instructions. Both resources are publicly available, paving the way for better integration of LLMs in multi-omics analysis. The Biology-Instructions is publicly available at: https://github.com/hhnqqq/Biology-Instructions.
The sequential recommendation task involves predicting the items users will be interested in next based on their past interaction sequence. Recently, sequential recommender systems with generative retrieval have garnered significant attention. However, during training, these generative recommenders focus only on maximizing the prediction probability of the next target item in the temporal sequence, while neglecting awareness of diverse plausible potential items.Although introducing large language models (LLMs) with world knowledge and adding a set of auxiliary tasks that can link item identifiers to their real-world meanings can alleviate this issue, the high inference costs associated with these LLM-based recommenders make them challenging to deploy in practical scenarios. In this paper, we propose a novel learning framework, LOHRec, which leverages the order and hierarchy in generative recommendation using quantized identifiers to further explore the performance ceiling of lightweight generative recommenders. Under fair comparisons with approximate backbone parameter sizes, comprehensive experiments show that all variants of generative recommenders using our framework outperform strong prior baselines across multiple datasets. Furthermore, we empirically demonstrate that LOHRec can efficiently align lightweight generative recommenders with LLM recommendation preferences in low-resource scenarios, further demonstrating its practical utility. Our code repository is available at [https://github.com/xjw-nlp/LOHRec](https://github.com/xjw-nlp/LOHRec).
Detecting bias in media content is crucial for maintaining information integrity and promoting inclusivity. Traditional methods analyze text from the writer’s perspective, which analyzes textual features directly from the writer’s intent, leaving the reader’s perspective underexplored. This paper investigates whether Large Language Models (LLMs) can be leveraged as readers for bias detection by generating reader-perspective comments. Experiments are conducted on the BASIL (news bias) and BeyondGender (gender bias) datasets with LLMs Gemma-7B, Phi-3-3.8B, Llama3.1-8B, Llama3.1-70B, and GPT4. The results demonstrate the effectiveness of reader-perspective comments for open-source LLMs, achieving performance comparable to GPT4’s. The findings highlight the significance of emotion-related comments, which are generally more beneficial than value-related ones in bias detection. In addition, experiments on Llamas show that comment selection ensures consistent performance regardless of model sizes and comment combinations. This study is particularly beneficial for small-size open-source LLMs.
Following Occam’s Razor: Dynamic Combination of Structured Knowledge for Multi-Hop Question Answering using LLMs
PDF ↗Multi-hop question answering is a challenging task that requires capturing information from different positions in multiple documents. Recently, several methods propose to enhance Large Language Models (LLMs) by incorporating structured knowledge, aiming to grasp key information for solving this task. Despite certain achievements, they still face the following challenges: 1) The neglect of text-based reasoning capabilities. 2) Information redundancy between text and triples. 3) Information loss during structured knowledge extraction. To solve the above challenges, in this paper, we propose Dynamic Combination of Structured Knowledge (DCSK), a novel framework for integrating text-based and triple-based paradigms. Following Occam’s Razor, DCSK dynamically determine the necessity of structured knowledge by the designed multi-faceted evaluation, which systematically assess the correctness, clarity, and informativeness of text-based prediction. For questions that require structured knowledge, we develop an iterative fact refiner that screens for question-relevant triples, verifies their factual adequacy, and thereby effectively excludes irrelevant and redundant information. Furthermore, based on the verification, we construct an adaptive knowledge reasoner that dynamically adjusts the need for text supplementation, thus mitigating the information deficiency in selected triples. Extensive experiments on three MHQA datasets demonstrate the efficiency and effectiveness of DCSK.
Generation-Augmented Retrieval: Rethinking the Role of Large Language Models in Zero-Shot Relation Extraction
PDF ↗Recent advances in Relation Extraction (RE) emphasize Zero-Shot methodologies, aiming to recognize unseen relations between entities with no annotated data. Although Large Language Models (LLMs) have demonstrated outstanding performance in many NLP tasks, their performance in Zero-Shot RE (ZSRE) without entity type constraints still lags behind Small Language Models (SLMs). LLM-based ZSRE often involves manual interventions and significant computational overhead, especially when scaling to large-scale multi-choice data.To this end, we introduce RE-GAR-AD, which not only leverages the generative capability of LLMs but also utilizes their representational power without tuning LLMs. We redefine LLM-based ZSRE as a retrieval challenge, utilizing a Generation-Augmented Retrieval framework coupled with a retrieval Adjuster. Specifically, our approach guides LLMs through crafted prompts to distill sentence semantics and enrich relation labels. We encode sentences and relation labels using LLMs and match their embeddings in a triplet fashion. This retrieval technique significantly reduces token input requirements. Additionally, to further optimize embeddings, we propose a plug-in retrieval adjuster with only 2M parameters, which allows rapid fine-tuning without accessing LLMs’ parameters. Our LLM-based model demonstrates comparable performance on multiple benchmarks.
This paper defines and explores the design space for information extraction (IE) from layout-rich documents using large language models (LLMs). The three core challenges of layout-aware IE with LLMs are 1) data structuring, 2) model engagement, and 3) output refinement. Our study investigates the sub-problems and methods within these core challenges, such as input representation, chunking, prompting, selection of LLMs, and multimodal models. It examines the effect of different design choices through LayIE-LLM, a new, open-source, layout-aware IE test suite, benchmarking against traditional, fine-tuned IE models. The results on two IE datasets show that LLMs require adjustment of the IE pipeline to achieve competitive performance: the optimized configuration found with LayIE-LLM achieves 13.3–37.5 F1 points more than a general-practice baseline configuration using the same LLM. To find a well-working configuration, we develop a one-factor-at-a-time (OFAT) method that achieves near-optimal results. Our method is only 0.8–1.8 points lower than the best full factorial exploration with a fraction (~2.8%) of the required computation. Overall, we demonstrate that, if well-configured, general-purpose LLMs match the performance of specialized models, providing a cost-effective, finetuning-free alternative. Our test-suite is available at https://github.com/gayecolakoglu/LayIE-LLM
Large Language Model Agents in Finance: A Survey Bridging Research, Practice, and Real-World Deployment
PDF ↗Large language models (LLMs) are increasingly applied to finance, yet challenges remain in aligning their capabilities with real-world institutional demands. In this survey, we provide a systematic, dual-perspective review bridging financial practice and LLM research. From a practitioner-centric standpoint, we introduce a functional taxonomy covering five core financial domains—Data Analysis, Investment Research, Trading, Investment Management, and Risk Management—mapping each to representative tasks, datasets, and institutional constraints. From a research-focused perspective, we analyze key modeling challenges, including numerical reasoning limitations, prompt sensitivity, and lack of real-time adaptability. We comprehensively catalog over 30 financial benchmarks and 20 representative models, and compare them across modalities, tasks, and deployment limitations. Finally, we identify open challenges and outline emerging directions such as continual adaptation, coordination-aware multi-agent systems, and privacy-compliant deployment. We emphasize deeper researcher–practitioner collaboration and transparent model architectures as critical pathways to safer and more scalable AI adoption in finance.
Factual completeness is a general term that captures how detailed and informative a factually correct text is. For instance, the factual sentence “Barack Obama was born in the United States” is factually correct, though less informative than the factual sentence “Barack Obama was born in Honolulu, Hawaii, United States”. Despite the known fact that LLMs tend to hallucinate and generate factually incorrect text, they might also tend to choose to generate factual text that is indeed factually correct and yet less informative than other, more informative choices. In this work, we tackle this problem by proposing an informativeness alignment mechanism. This mechanism takes advantage of recent factual informativeness benchmarks to propose an informativeness alignment objective. This objective prioritizes answers that are both correct and informative. We find that when training a model to maximize this objective or optimize its preference, we can improve not just informativeness but also factuality.
A Comprehensive Survey on Learning from Rewards for Large Language Models: Reward Models and Learning Strategies
PDF ↗Recent developments in Large Language Models (LLMs) have shifted from pre-training scaling to post-training and test-time scaling. Across these developments, a key unified paradigm has arisen: Learning from Rewards, where reward signals act as the guiding stars to steer LLM behavior. It has underpinned a wide range of prevalent techniques, such as reinforcement learning (RLHF, RLAIF, DPO, and GRPO), reward-guided decoding, and post-hoc correction. Crucially, this paradigm enables the transition from passive learning from static data to active learning from dynamic feedback. This endows LLMs with aligned preferences and deep reasoning capabilities for diverse tasks. In this survey, we present a comprehensive overview of learning from rewards, from the perspective of reward models and learning strategies across training, inference, and post-inference stages. We further discuss the benchmarks for reward models and the primary applications. Finally we highlight the challenges and future directions.
Language models can be steered by modifying their internal representations to control concepts such as emotion, style, or truthfulness in generation. However, the conditions for an effective intervention remain unclear and are often validated through heuristics and trial-and-error. To fill this gap, we demonstrate that intervention efficacy, measured by linear steerability (i.e., the ability to adjust output via linear transformations of hidden states), emerges during intermediate stages of training. Moreover, even closely related concepts (e.g., anger and sadness) exhibit steerability emergence at distinct stages of training*.To better interpret the dynamics of steerability during training, we adapt existing intervention techniques into a unified framework, referred to as the “Intervention Detector” (ID), which is designed to reveal how linear steerability evolves over the course of training through hidden state and representation analysis. ID reveals that concepts become increasingly linearly separable in the hidden space as training progresses, which strongly correlates with the emergence of linear steerability. We further introduce ID-based metrics, such as heatmaps, entropy trends, and cosine similarity, to help interpret how linear steerability evolves throughout training. In addition, we apply ID across different model families to ensure the generality of our findings on steerability dynamics.
Semantic-Aware Action Space Compression via LLM-DRL Synergy for Efficient Task-oriented Dialogue Policy Exploration
PDF ↗The flexibility of natural language significantly expands the action space in task-oriented dialogue systems, causing inefficient exploration and slow convergence in deep reinforcement learning (DRL)-based policy optimization. Pre-trained large language models (LLMs), with world knowledge and semantic understanding, offer promising solutions. To this end, we propose LLM-Guided DRL via Semantic-Aware Action Pruning (LLMSAP), a novel framework that synergizes pretrained LLMs with DRL. LLMSAP leverages the world knowledge and contextual understanding of LLMs to guide decision-making via an action feasibility assessment. Instead of requiring LLMs to directly generate optimal actions due to their limited precision in sequential decision tasks, LLMSAP employs a lightweight action pruning mechanism. Specifically, LLMs act as action filters, rapidly eliminating semantically implausible or low-potential actions from multi-turn dialogue context, allowing the DRL agent to focus exploration on a refined candidate subset. This two-stage framework (“prune-then-optimize”) avoids extensive LLM fine-tuning while preserving the decision-making precision of DRL. Experiments on multiple benchmarks verify the effectiveness of LLMSAP.
We investigate the effect of duplicating multihead self-attention layers in large language models (LLMs) across a range of language tasks, with and without fine-tuning. The results demonstrate that duplicating the initial layers once or twice often yields a significant performance boost. Attention analysis uncovered the underlying mechanisms driving the improvement when performing layer duplication. This method enhances LLM capabilities with or without additional training or labeled data.
Large Language Models (LLMs) exhibit significant potential in complex software engineering tasks, however, their fault localization capabilities within repository are constrained by inherent limitations in max context length. Although Test-Time Scaling (TTS) can generate multiple candidate solutions, traditional selection strategies often fail to identify the optimal one. To solve this problem, we introduces Hierarchical Localization Reward Model (HiLoRM), which specifically designed to evaluate and select the most accurate fault localization candidates (at file, function, and line levels) from the multiple sampled outputs of LLMs, thereby enhancing localization accuracy. Furthermore, we constructed the HiFL-44k dataset, comprising approximately 44,000 fault localization instances, to train HiLoRM. Experimental results demonstrate that on the SWE-Bench-Lite dataset, HiLoRM improves the final line-level localization recall by 12% compared to a baseline model that does not use a reward model. Concurrently, HiLoRM exhibits a strong capability to evaluate predictions from larger LLMs (e.g., 32B parameters) and demonstrates transferability and generalization potential when applied to other fault localization methods. This work provides an effective methodology and an accessible model to significantly improve the accuracy and reliability of LLMs for repository-level fault localization. Our codes and datasets are available at https://github.com/SZU-ZJW/HiFL-Method.