The training of large language models (LLMs) is expensive. In this paper, we study data-efficient approaches for pre-training LLMs, i.e., techniques that aim to optimize the Pareto frontier of model quality and training resource/data consumption. We seek to understand the tradeoffs associated with data selection routines based on (i) expensive-to-compute data-quality estimates, and (ii) maximization of coverage and diversity-based measures in the feature space. Our first technique, Ask-LLM, leverages the zero-shot reasoning capabilities of instruction-tuned LLMs to directly assess the quality of a training example. To target coverage, we propose Density sampling, which models the data distribution to select a diverse sample. In our comparison of 19 samplers, involving hundreds of evaluation tasks and pre-training runs, we find that Ask-LLM and Density are the best methods in their respective categories. Coverage sampling can recover the performance of the full data, while models trained on Ask-LLM data consistently outperform full-data training -- even when we reject 90% of the original dataset, while converging up to 70% faster.

Evaluating disfluency removal in speech requires more than aggregate token-level scores. Traditional word-based metrics such as precision, recall, and F1 (E-Scores) capture overall performance but cannot reveal why models succeed or fail. We introduce Z-Scores, a span-level linguistically-grounded evaluation metric that categorizes system behavior across distinct disfluency types (EDITED, INTJ, PRN). Our deterministic alignment module enables robust mapping between generated text and disfluent transcripts, allowing Z-Scores to expose systematic weaknesses that word-level metrics obscure. By providing category-specific diagnostics, Z-Scores enable researchers to identify model failure modes and design targeted interventions – such as tailored prompts or data augmentation – yielding measurable performance improvements. A case study with LLMs shows that Z-Scores uncover challenges with INTJ and PRN disfluencies hidden in aggregate F1, directly informing model refinement strategies.

LLM powered multi-agent systems (MAS) have emerged as a promising paradigm for complex tasks. However, their advantages over single-agent systems (SAS) remain unclear, with performance varying inconsistently across settings. Here, we provide an information bottleneck perspective on elucidating the differences between MAS and SAS. Specifically, our key observation is that a SAS accumulates its full reasoning trace in one shared context, while a MAS uses isolated local contexts connected by bounded relay messages. We show that, under infinite relay bandwidth, any SAS can be simulated by a MAS that transmits the full upstream context. Thus, the nontrivial advantage of MAS arises under bounded relays, where compression introduces a fundamental trade-off: reducing redundant context can improve efficiency, but may also incur loss of task-relevant information. We formalize this trade-off as an information bottleneck controlled by an effective parameter β, which captures how the balance shifts with model capability, and shows that MAS gains arise when context reduction outweighs relay information loss. We conduct 18 controlled experiments across five benchmarks and three model scales to validate our theoretical studies. We observe that MAS consistently helps when relays are near-sufficient, especially for weaker models. In contrast, MAS gains shrink or reverse when relays incur information loss, especially for stronger models that can already extract useful information from redundant context and thus gain little from compression. Our study shows that multi-agent design is fundamentally an information-bottleneck optimization problem. This perspective explains when bounded inter-agent communication helps or hurts.

Prompting is central to interaction with AI systems, yet many users struggle to explore alternative directions, articulate creative intent, or understand how variations in prompts shape model outputs. We introduce prompt recommender systems (PRS) as an interaction approach that supports exploration, suggesting contextually relevant follow-up prompts. We present PromptHelper, a PRS prototype integrated into an AI chatbot that surfaces semantically diverse prompt suggestions while users work on real writing tasks. We evaluate PromptHelper in a 2x2 fully within-subjects study (N = 32) across creative and academic writing tasks. Results show that PromptHelper significantly increases users’ perceived exploration and expressiveness without increasing cognitive workload. Qualitative findings illustrate how prompt recommendations help users branch into new directions, overcome uncertainty about what to ask next, and better articulate their intent. We release open-source resources to support research on prompt recommendation. Our findings point to future work on AI interfaces that scaffold exploratory interaction while preserving user agency.

Effective and efficient access to relevant information is essential for disaster management. However, no retrieval model is specialized for disaster management, and existing general-domain models fail to handle the varied search intents inherent to disaster management scenarios, resulting in inconsistent and unreliable performance. To this end, we introduce DMRetriever, the first series of dense retrieval models (33M to 7.6B) tailored for this domain. It is trained through a novel three-stage framework of bidirectional attention adaptation, unsupervised contrastive pre-training, and difficulty-aware progressive instruction fine-tuning, using high-quality data generated through an advanced data refinement pipeline. Comprehensive experiments demonstrate that DMRetriever achieves state-of-the-art performance across all six search intents at every model scale. Moreover, DMRetriever is highly parameter-efficient, with 596M model outperforming baselines over 13.3 larger and 33M model exceeding baselines with only 7.6% of their parameters. All codes, data, and checkpoints are available at https://github.com/KaiYin97/DMRETRIEVER.

Speech Large Language Models (SpeechLLMs) process spoken input directly, retaining cues such as accent and perceived gender that were previously removed in cascaded pipelines. This introduces speaker identity dependent variation in responses. We present a large-scale intersectional evaluation of accent and gender bias in three SpeechLLMs using 2,880 controlled interactions across six English accents and two gender presentations, keeping linguistic content constant through voice cloning. Using pointwise LLM-judge ratings, pairwise comparisons, and Best-Worst Scaling with human validation, we detect consistent disparities. Eastern European-accented speech receives lower helpfulness scores, particularly for female-presenting voices. The bias is implicit: responses remain polite but differ in helpfulness. While LLM judges capture the directional trend of these biases, human evaluators exhibit significantly higher sensitivity, uncovering sharper intersectional disparities.

Accurate question answering (QA) in disaster management requires reasoning over uncertain and conflicting information, a setting poorly captured by existing benchmarks built on clean evidence. We introduce DisastQA, a large-scale benchmark of 3,000 rigorously verified questions (2,000 multiple-choice and 1,000 open-ended) spanning eight disaster types. The benchmark is constructed via a human-LLM collaboration pipeline with stratified sampling to ensure balanced coverage. Models are evaluated under varying evidence conditions, from closed-book to noisy evidence integration, enabling separation of internal knowledge from reasoning under imperfect information. For open-ended QA, we propose a human-verified keypoint-based evaluation protocol emphasizing factual completeness over verbosity. Experiments with 20 models reveal substantial divergences from general-purpose leaderboards such as MMLU-Pro. While recent open-weight models approach proprietary systems in clean settings, performance degrades sharply under realistic noise, exposing critical reliability gaps for disaster response. All code, data, and evaluation resources are available at https://github.com/TamuChen18/DisastQA_open.

Sequential recommenders are typically trained on long user histories to capture rich behavioral signals, yet serving with training-length sequences is often impractical due to real-time efficiency constraints. Directly using only recent behaviors leads to a severe performance drop. To bridge this gap, existing approaches compress user histories into persistent per-user states, storing and retrieving them at inference time; while effective, they impose non-trivial infrastructure overhead and offer little remedy in cold-start scenarios. In this paper, we empirically identify two structural flaws rooted in geometric properties and dataset sparsity, and propose a novel two-stage framework to close the long-short-view performance gap. Specifically, in the first stage, we replace the commonly used dot-product with angular similarity scoring and leverage a modified softmax to counter prefix position bias. In the second stage, we fine-tune only bias and LayerNorm components, which are universal to standard sequential backbones, for further improvement. Both stages are guided by carefully designed learning objectives. Extensive experiments on two representative backbones across three public datasets demonstrate the effectiveness of our proposed framework.

Automatic speech recognition (ASR) evaluation compares system output to ground truth transcripts, with Word Error Rate (WER) quantifying the distance between them. But ground truth transcripts are not discovered - they are produced by human annotators following conventions that encode normative assumptions about which speech features matter. Different conventions (verbatim, non-verbatim, legal) produce different transcripts of identical speech and judge the same ASR output differently. This paper argues that reference monism - enforcing a single transcription convention as ground truth - commits epistemic injustice. Speakers with aphasia, whose speech includes clinically meaningful disfluencies, are systematically disadvantaged when evaluated against "clean" references that treat those disfluencies as errors. The harm is not merely differential performance, but that evaluative infrastructure lacks interpretive resources to recognize their contributions as legitimate. We develop a philosophical framework introducing the hermeneutical gap, formalize Epistemic Injustice Distance (EID) to measure reference monism's cost, and demonstrate empirically using AphasiaBank that WER varies depending on which convention defines ground truth. We propose WER-Range: reporting performance across legitimate conventions rather than assuming a single correct answer.

Disasters cause severe societal impacts, demanding rapid coordination of heterogeneous AI tools, from satellite analysis to flood prediction and damage assessment, into coherent multi-step workflows. As LLMs increasingly serve as orchestrators of such pipelines, effective coordination requires more than selecting semantically plausible tools: LLMs must generate executable workflows with correct parameter binding and dependency propagation. We introduce DisasterBench, a benchmark for evaluating structured multi-agent planning over semantically similar but operationally distinct disaster-response tools. To enable step-level failure attribution, we further propose First-Point-of-Failure (FPoF), which localizes the earliest root cause in a predicted workflow, separating primary errors from downstream cascading effects. Our evaluation reveals three findings: planning method effectiveness depends strongly on model capacity; tool mismatch and parameter-binding errors dominate first failures, revealing semantic grounding and execution consistency as distinct bottlenecks; and verbose intermediate reasoning can create instruction clash with structured output requirements, disrupting plan generation. Together, these findings highlight a fundamental gap between semantic reasoning and execution-grounded coordination, underscoring the need for planning frameworks that jointly model semantic intent, execution constraints, and workflow consistency. Code, data, and evaluation resources are available at: https://github.com/TamuChen18/DisasterBench_Open

Automated methods for the assessment of replicability of scientific claims offer a scalable complement to replication studies and traditional peer review. Drawing on a large dataset of claims, human judgments, and a limited set of replication outcomes, we developed and evaluated three distinct artificial intelligence systems designed to predict human expert assessments of replicability using diverse methodologies—including synthetic prediction markets, interpretable feature-based modeling, knowledge graph reasoning, and semantic parsing with argument structures. While these systems achieved modest calibration to human judgment distributions, they failed to discriminate between replicable and non-replicable claims. Our findings suggest that while machine assessments of research replicability may complement human reasoning, their current performance limitations and opportunities for bias demand careful evaluation before real-world application.

We study the transformation of autoregressive models (ARLMs) into diffusion language models (DLMs). Rather than pretraining from scratch, prior work replaces the causal attention in ARLMs with bidirectional attention and then trains the resulting model using a DLM objective. However, these approaches incur two distribution shifts. First, transitioning from a next-token prediction objective to a DLM objective can discard knowledge acquired by the ARLM during training. Second, standard DLMs suffer from a train-inference mismatch, as the training loss is defined on randomly masked sequences rather than the trajectories encountered at inference produced by confidence-based decoding. To address both challenges, we introduce an On-Policy Diffusion Language Model (OPDLM) in which On-Policy Distillation (OPD) is employed for ARLM-to-DLM transformation. Specifically, OPDLM is trained via self-OPD, where the student, an ARLM with bidirectional attention, generates its own trajectories, and the teacher, the original frozen ARLM, distills its knowledge by providing target logits on these trajectories. By training directly in an on-policy manner, OPDLM eliminates the train-inference mismatch in DLMs, while distillation from the original model enhances knowledge retention from the ARLM. Empirical results demonstrate that OPDLM requires 15x to 7,000x fewer training tokens with strong performance across a wide variety of tasks. OPDLM avoids the prohibitive cost of DLM pretraining and positions DLM transformation as a form of ARLM post-training.

Persona-assigned Large Language Models can adopt diverse roles, enabling personalized and context-aware reasoning. However, even minor demographic perturbations in personas, such as simple pronoun swaps, can alter reasoning trajectories, leading to divergent sets of correct answers on reasoning benchmarks. We explore the potential of these variations as a constructive resource to improve LLM reasoning performance. We propose CHOIR (Collaborative Harmonization fOr Inference Robustness), a test-time framework that harmonizes a set of demographically perturbed, persona-conditioned reasoning signals into a unified prediction. CHOIR orchestrates a collaborative decoding process among counterfactual personas perturbed across dimensions of gender, race, religion, disability, and age, dynamically balancing agreement and divergence in their reasoning paths to improve performance. Experiments demonstrate that CHOIR consistently enhances LLM reasoning across model architectures, scales, and tasks. Improvements reach up to 20.1% for individual groups and 15.1% on average, and we show that CHOIR remains effective even when base personas are suboptimal.

Current speech translation systems, including SpeechLLMs, are trained on cleaned text and tend to strip disfluencies like filled pauses and false starts rather than translate them. We show this comes at a cost: disfluencies carry meaning that gets lost when speech is cleaned up. To study this systematically, we introduce Uh-Mazing, a benchmark of human-translated, disfluency-annotated Switchboard speech covering English into eight target languages. Across these languages and several architectures, we find that false starts and self-repairs, not filled pauses or discourse markers, drive most of the translation-quality loss, and that models which fail to preserve a disfluency tend to omit it rather than mistranslate it. We show inference-time decoding can mitigate this without retraining, and release the benchmark and code.

Large Language Models (LLMs) excel at capturing latent semantics and contextual relationships across diverse modalities. However, in modeling user behavior from sequential interaction data, performance often suffers when such semantic context is limited or absent. We introduce LaMAR, a LLM-driven semantic enrichment framework designed to enrich such sequences automatically. LaMAR leverages LLMs in a few-shot setting to generate auxiliary contextual signals by inferring latent semantic aspects of a user's intent and item relationships from existing metadata. These generated signals, such as inferred usage scenarios, item intents, or thematic summaries, augment the original sequences with greater contextual depth. We demonstrate the utility of this generated resource by integrating it into benchmark sequential modeling tasks, where it consistently improves performance. Further analysis shows that LLM-generated signals exhibit high semantic novelty and diversity, enhancing the representational capacity of the downstream models. This work represents a new data-centric paradigm where LLMs serve as intelligent context generators, contributing a new method for the semi-automatic creation of training data and language resources.

Understanding how online communities discuss and make sense of complex social issues is a central challenge in social media research, yet existing tools for large-scale discourse analysis are often closed-source, difficult to adapt, or limited to single analytical views. We present SocialPulse, an open-source subreddit sensemaking toolkit that unifies multiple complementary analyses – topic modeling, sentiment analysis, user activity characterization, and bot detection – within a single interactive system. SocialPulse enables users to fluidly move between aggregate trends and fine-grained content, compare highly active and long-tail contributors, and examine temporal shifts in discourse across subreddits. The demo showcases end-to-end exploratory workflows that allow researchers and practitioners to rapidly surface themes, participation patterns, and emerging dynamics in large Reddit datasets. By offering an extensible and openly available platform, SocialPulse provides a practical and reusable foundation for transparent, reproducible sensemaking of online community discourse.

We aim to improve the reasoning capabilities of diffusion language models (DLMs). While SFT performs well for autoregressive models, its use in DLMs faces challenges. Our observation and analysis reveal that vanilla SFT does not consider learnability, i.e., what and when tokens are learned. Specifically, we observe that rare tokens are difficult to learn when most of the input is masked. In contrast, it is straightforward and thus of little value to learn common tokens when most of the input is unmasked. To consider learnability, we propose LIFT, a learnability-informed fine-tuning strategy for DLMs. LIFT learns easy tokens when most of the input is noisy and hard tokens when more input is available, thereby aligning training with information available at different diffusion time steps. Our results show that LIFT outperforms existing SFT baselines across six reasoning benchmarks, achieving up to a 3x relative gain on AIME'24 and AIME'25.