arXiv subjects

Jennifer Neville

Publications and source records attributed to Jennifer Neville.

At least 19 recordsLinked to original sources

arXiv2026

LLMs Get Lost in Evolving User Intent

As LLMs become more capable, they are increasingly deployed as collaborative agents, taking on user-delegated tasks through iterative interaction. Yet genuine interaction is inherently dynamic: users rarely specify their intent upfront, instead disclosing, revising, and reshaping it as the conversation unfolds. Despite this, LLMs are still predominantly evaluated or trained in single-turn, fully-specified settings, leaving open a fundamental question: how well do LLMs track and act on user intent as it evolves over the course of a conversation? To study this, we introduce a framework that transforms static, single-turn tasks into dynamic multi-turn conversations in which the user's intent evolves across turns--incrementally revealed, revised, and at times redirected mid-conversation--while preserving each task's original evaluation protocol, enabling existing benchmarks to be reused as controlled testbeds without new annotation. Across multiple tasks, we surface a consistent phenomenon: strong static-setting performance does not transfer to the evolving-intent setting, with substantial drops across model families. Our findings point to a fundamental gap: today's LLMs do not yet faithfully track and act on the user's evolving intent, a capability invisible to static evaluation yet critical for future collaborative agents.

cs.LG↗

arXiv2026

Generating Leakage-Free Benchmarks for Robust RAG Evaluation

Retrieval-augmented generation (RAG) is widely used to augment large language models (LLMs) with external knowledge. However, many benchmark datasets, designed to test RAG performance, comprise many questions that can already be answered from an LLM's parametric memory. This leads to unreliable evaluation. We refer to this phenomenon as knowledge leakage: cases where RAG tasks are solvable without retrieval. This issue worsens over time due to benchmark aging. As benchmarks are reused for training, their contents are increasingly absorbed into model parameters, making them less effective for evaluating retrieval. We introduce SeedRG, a semi-synthetic benchmark generation pipeline that mitigates knowledge leakage and addresses the issue of benchmark aging. Starting from a seed benchmark dataset, SeedRG extracts a reasoning graph from question-context pairs to capture their underlying reasoning structure, and then generates new examples via type-constrained entity replacement. This process produces structurally similar but novel instances that are unlikely to exist in the model's parametric knowledge, while preserving the original reasoning patterns. To ensure quality, we incorporate two verification steps: (1) a reasoning-graph consistency check to maintain task difficulty, and (2) a knowledge-leakage filter to exclude instances answerable without retrieval.

cs.CL↗

arXiv2026

LLMs Corrupt Your Documents When You Delegate

Large Language Models (LLMs) are poised to disrupt knowledge work, with the emergence of delegated work as a new interaction paradigm (e.g., vibe coding). Delegation requires trust - the expectation that the LLM will faithfully execute the task without introducing errors into documents. We introduce DELEGATE-52 to study the readiness of AI systems in delegated workflows. DELEGATE-52 simulates long delegated workflows that require in-depth document editing across 52 professional domains, such as coding, crystallography, and music notation. Our large-scale experiment with 19 LLMs reveals that current models degrade documents during delegation: even frontier models (Gemini 3.1 Pro, Claude 4.6 Opus, GPT 5.4) corrupt an average of 25% of document content by the end of long workflows, with other models failing more severely. Additional experiments reveal that agentic tool use does not improve performance on DELEGATE-52, and that degradation severity is exacerbated by document size, length of interaction, or presence of distractor files. Our analysis shows that current LLMs are unreliable delegates: they introduce sparse but severe errors that silently corrupt documents, compounding over long interaction.

cs.CL↗

arXiv2026

Response-Aware User Memory Selection for LLM Personalization

A common approach to personalization in large language models (LLMs) is to incorporate a subset of the user memory into the prompt at inference time to guide the model's generation. Existing methods select these subsets primarily using similarity between user memory items and input queries, ignoring how features actually affect the model's response distribution. We propose Response-Utility optimization for Memory Selection (RUMS), a novel method that selects user memory items by measuring the mutual information between a subset of memory and the model's outputs, identifying items that reduce response uncertainty and sharpen predictions beyond semantic similarity. We demonstrate that this information-theoretic foundation enables more principled user memory selection that aligns more closely with human selection compared to state-of-the-art methods, and models $400\times$ larger. Additionally, we show that memory items selected using RUMS result in better response quality compared to existing approaches, while having up to $95\%$ reduction in computational cost.

cs.AI↗

arXiv2026

DP-RFT: Learning to Generate Synthetic Text via Differentially Private Reinforcement Fine-Tuning

Differentially private (DP) synthetic data generation plays a pivotal role in developing large language models (LLMs) on private data, where data owners cannot provide eyes-on access to individual examples. Generating DP synthetic data typically involves a difficult trade-off. On one hand, DP finetuning methods train an LLM as a synthetic data generator with formal privacy guarantees, yet it still requires the raw content of private examples for model training. However, methods that avoid direct exposure to private data are bounded by an off-the-shelf, un-finetuned model, whose outputs often lack domain fidelity. Can we train an LLM to generate high-quality synthetic text without eyes-on access to individual private examples? In this work, we introduce Differentially Private Reinforcement Fine-Tuning (DP-RFT), an online reinforcement learning algorithm for synthetic data generation with LLMs. DP-RFT leverages DP-protected nearest-neighbor votes from an eyes-off private corpus as a reward signal for on-policy synthetic samples generated by an LLM. The LLM iteratively learns to generate synthetic data to maximize the expected DP votes through Proximal Policy Optimization (PPO). We evaluate DP-RFT for long-form and domain-specific synthetic data generation, such as news articles, meeting transcripts, and medical article abstracts. Our experiments show that DP-RFT closes the gap between private evolution and DP finetuning methods in terms of the fidelity and downstream utility of the generated synthetic data, while respecting the private data boundary.

cs.CL↗

arXiv2026

Reasoning about Reasoning: BAPO Bounds on Chain-of-Thought Token Complexity in LLMs

Inference-time scaling via chain-of-thought (CoT) reasoning is a major driver of state-of-the-art LLM performance, but it comes with substantial latency and compute costs. We address a fundamental theoretical question: how many reasoning tokens are required to solve a problem as input size grows? By extending the bounded attention prefix oracle (BAPO) model--an abstraction of LLMs that quantifies the information flow required to solve a task--we prove lower bounds on the CoT tokens required for three canonical BAPO-hard tasks: binary majority, triplet matching, and graph reachability. We show that each requires $\Omega(n)$ reasoning tokens when the input size is $n$. We complement these results with matching or near-matching upper bounds via explicit constructions. Finally, our experiments with frontier reasoning models show approximately linear reasoning token scaling on these tasks and failures when constrained to smaller reasoning budgets, consistent with our theoretical lower bounds. Together, our results identify fundamental bottlenecks in inference-time compute through CoT and offer a principled tool for analyzing optimal reasoning length.

cs.AI↗

arXiv2025

Flipping the Dialogue: Training and Evaluating User Language Models

Conversations with LMs involve two participants: a human user leading the conversation, and an LM assistant responding to the user's request. To satisfy this specific role, LMs are post-trained to be helpful assistants -- optimized to produce exhaustive and well-structured responses, free of ambiguity and grammar errors. User utterances, on the other hand, are rarely perfected, with each user phrasing requests in unique ways, sometimes putting in partial effort at each turn and refining on the fly. To evaluate LM performance in realistic settings, prior work simulated users in multi-turn conversations, often by prompting an LM originally trained to be a helpful assistant to act as a user. However, we show that assistant LMs make for poor user simulators, with the surprising finding that better assistants yield worse simulators. Instead, we introduce purpose-built User Language Models (User LMs) - models post-trained to simulate human users in multi-turn conversations. Through various evaluations, we show how User LMs align better with human behavior and achieve better simulation robustness than existing simulation methods. When leveraging User LMs to simulate coding and math conversations, the performance of a strong assistant (GPT-4o) drops from 74.6% to 57.4%, confirming that more realistic simulation environments lead to assistant struggles as they fail to cope with the nuances of users in multi-turn setups.

cs.CL↗

arXiv2025

Lost in Transmission: When and Why LLMs Fail to Reason Globally

Despite their many successes, transformer-based large language models (LLMs) continue to struggle with tasks that require complex reasoning over large parts of their input. We argue that these failures arise due to capacity limits on the accurate flow of information within LLMs. To formalize this issue, we introduce the bounded attention prefix oracle (BAPO) model, a new computational framework that models bandwidth constraints on attention heads, the mechanism for internal communication in LLMs. We show that several important reasoning problems like graph reachability require high communication bandwidth for BAPOs to solve; we call these problems BAPO-hard. Our experiments corroborate our theoretical predictions: GPT-4o, Claude, and Gemini succeed on BAPO-easy tasks and fail even on relatively small BAPO-hard tasks. BAPOs also reveal another benefit of chain of thought (CoT): we prove that breaking down a task using CoT can turn any BAPO-hard problem into a BAPO-easy one. Our results offer principled explanations for key LLM failures and suggest directions for architectures and inference methods that mitigate bandwidth limits.

cs.AI↗

arXiv2025

LLMs Get Lost In Multi-Turn Conversation

Large Language Models (LLMs) are conversational interfaces. As such, LLMs have the potential to assist their users not only when they can fully specify the task at hand, but also to help them define, explore, and refine what they need through multi-turn conversational exchange. Although analysis of LLM conversation logs has confirmed that underspecification occurs frequently in user instructions, LLM evaluation has predominantly focused on the single-turn, fully-specified instruction setting. In this work, we perform large-scale simulation experiments to compare LLM performance in single- and multi-turn settings. Our experiments confirm that all the top open- and closed-weight LLMs we test exhibit significantly lower performance in multi-turn conversations than single-turn, with an average drop of 39% across six generation tasks. Analysis of 200,000+ simulated conversations decomposes the performance degradation into two components: a minor loss in aptitude and a significant increase in unreliability. We find that LLMs often make assumptions in early turns and prematurely attempt to generate final solutions, on which they overly rely. In simpler terms, we discover that *when LLMs take a wrong turn in a conversation, they get lost and do not recover*.

cs.CL↗

arXiv2025

Group Preference Alignment: Customized LLM Response Generation from In-Situ Conversations

LLMs often fail to meet the specialized needs of distinct user groups due to their one-size-fits-all training paradigm \cite{lucy-etal-2024-one} and there is limited research on what personalization aspects each group expect. To address these limitations, we propose a group-aware personalization framework, Group Preference Alignment (GPA), that identifies context-specific variations in conversational preferences across user groups and then steers LLMs to address those preferences. Our approach consists of two steps: (1) Group-Aware Preference Extraction, where maximally divergent user-group preferences are extracted from real-world conversation logs and distilled into interpretable rubrics, and (2) Tailored Response Generation, which leverages these rubrics through two methods: a) Context-Tuned Inference (GAP-CT), that dynamically adjusts responses via context-dependent prompt instructions, and b) Rubric-Finetuning Inference (GPA-FT), which uses the rubrics to generate contrastive synthetic data for personalization of group-specific models via alignment. Experiments demonstrate that our framework significantly improves alignment of the output with respect to user preferences and outperforms baseline methods, while maintaining robust performance on standard benchmarks.

cs.CL↗

arXiv2025

GenTool: Enhancing Tool Generalization in Language Models through Zero-to-One and Weak-to-Strong Simulation

Large Language Models (LLMs) can enhance their capabilities as AI assistants by integrating external tools, allowing them to access a wider range of information. While recent LLMs are typically fine-tuned with tool usage examples during supervised fine-tuning (SFT), questions remain about their ability to develop robust tool-usage skills and can effectively generalize to unseen queries and tools. In this work, we present GenTool, a novel training framework that prepares LLMs for diverse generalization challenges in tool utilization. Our approach addresses two fundamental dimensions critical for real-world applications: Zero-to-One Generalization, enabling the model to address queries initially lacking a suitable tool by adopting and utilizing one when it becomes available, and Weak-to-Strong Generalization, allowing models to leverage enhanced versions of existing tools to solve queries. To achieve this, we develop synthetic training data simulating these two dimensions of tool usage and introduce a two-stage fine-tuning approach: optimizing tool ranking, then refining tool selection. Through extensive experiments across four generalization scenarios, we demonstrate that our method significantly enhances the tool-usage capabilities of LLMs ranging from 1B to 8B parameters, achieving performance that surpasses GPT-4o. Furthermore, our analysis also provides valuable insights into the challenges LLMs encounter in tool generalization.

cs.CL↗

arXiv2024

Rethinking Node Representation Interpretation through Relation Coherence

Understanding node representations in graph-based models is crucial for uncovering biases ,diagnosing errors, and building trust in model decisions. However, previous work on explainable AI for node representations has primarily emphasized explanations (reasons for model predictions) rather than interpretations (mapping representations to understandable concepts). Furthermore, the limited research that focuses on interpretation lacks validation, and thus the reliability of such methods is unclear. We address this gap by proposing a novel interpretation method-Node Coherence Rate for Representation Interpretation (NCI)-which quantifies how well different node relations are captured in node representations. We also propose a novel method (IME) to evaluate the accuracy of different interpretation methods. Our experimental results demonstrate that NCI reduces the error of the previous best approach by an average of 39%. We then apply NCI to derive insights about the node representations produced by several graph-based methods and assess their quality in unsupervised settings.

cs.LG↗

arXiv2024

Improving Node Representation by Boosting Target-Aware Contrastive Loss

Graphs model complex relationships between entities, with nodes and edges capturing intricate connections. Node representation learning involves transforming nodes into low-dimensional embeddings. These embeddings are typically used as features for downstream tasks. Therefore, their quality has a significant impact on task performance. Existing approaches for node representation learning span (semi-)supervised, unsupervised, and self-supervised paradigms. In graph domains, (semi-)supervised learning often only optimizes models based on class labels, neglecting other abundant graph signals, which limits generalization. While self-supervised or unsupervised learning produces representations that better capture underlying graph signals, the usefulness of these captured signals for downstream target tasks can vary. To bridge this gap, we introduce Target-Aware Contrastive Learning (Target-aware CL) which aims to enhance target task performance by maximizing the mutual information between the target task and node representations with a self-supervised learning process. This is achieved through a sampling function, XGBoost Sampler (XGSampler), to sample proper positive examples for the proposed Target-Aware Contrastive Loss (XTCL). By minimizing XTCL, Target-aware CL increases the mutual information between the target task and node representations, such that model generalization is improved. Additionally, XGSampler enhances the interpretability of each signal by showing the weights for sampling the proper positive examples. We show experimentally that XTCL significantly improves the performance on two target tasks: node classification and link prediction tasks, compared to state-of-the-art models.

cs.LG↗

arXiv2024

WildFeedback: Aligning LLMs With In-situ User Interactions And Feedback

As large language models (LLMs) continue to advance, aligning these models with human preferences has emerged as a critical challenge. Traditional alignment methods, relying on human or LLM annotated datasets, are limited by their resource-intensive nature, inherent subjectivity, misalignment with real-world user preferences, and the risk of feedback loops that amplify model biases. To overcome these limitations, we introduce WildFeedback, a novel framework that leverages in-situ user feedback during conversations with LLMs to create preference datasets automatically. Given a corpus of multi-turn user-LLM conversation, WildFeedback identifies and classifies user feedback to LLM responses between conversation turns. The user feedback is then used to create examples of preferred and dispreferred responses according to users' preference. Our experiments demonstrate that LLMs fine-tuned on WildFeedback dataset exhibit significantly improved alignment with user preferences, as evidenced by both traditional benchmarks and our proposed checklist-guided evaluation. By incorporating in-situ feedback from actual users, WildFeedback addresses the scalability, subjectivity, and bias challenges that plague existing approaches, marking a significant step toward developing LLMs that are more responsive to the diverse and evolving needs of their users.

cs.CL↗

arXiv2024

LLMExplainer: Large Language Model based Bayesian Inference for Graph Explanation Generation

Recent studies seek to provide Graph Neural Network (GNN) interpretability via multiple unsupervised learning models. Due to the scarcity of datasets, current methods easily suffer from learning bias. To solve this problem, we embed a Large Language Model (LLM) as knowledge into the GNN explanation network to avoid the learning bias problem. We inject LLM as a Bayesian Inference (BI) module to mitigate learning bias. The efficacy of the BI module has been proven both theoretically and experimentally. We conduct experiments on both synthetic and real-world datasets. The innovation of our work lies in two parts: 1. We provide a novel view of the possibility of an LLM functioning as a Bayesian inference to improve the performance of existing algorithms; 2. We are the first to discuss the learning bias issues in the GNN explanation problem.

cs.LG↗

arXiv2024

On Overcoming Miscalibrated Conversational Priors in LLM-based Chatbots

We explore the use of Large Language Model (LLM-based) chatbots to power recommender systems. We observe that the chatbots respond poorly when they encounter under-specified requests (e.g., they make incorrect assumptions, hedge with a long response, or refuse to answer). We conjecture that such miscalibrated response tendencies (i.e., conversational priors) can be attributed to LLM fine-tuning using annotators -- single-turn annotations may not capture multi-turn conversation utility, and the annotators' preferences may not even be representative of users interacting with a recommender system. We first analyze public LLM chat logs to conclude that query under-specification is common. Next, we study synthetic recommendation problems with configurable latent item utilities and frame them as Partially Observed Decision Processes (PODP). We find that pre-trained LLMs can be sub-optimal for PODPs and derive better policies that clarify under-specified queries when appropriate. Then, we re-calibrate LLMs by prompting them with learned control messages to approximate the improved policy. Finally, we show empirically that our lightweight learning approach effectively uses logged conversation data to re-calibrate the response strategies of LLM-based chatbots for recommendation tasks.

cs.IR↗

arXiv2024

MS MARCO Web Search: a Large-scale Information-rich Web Dataset with Millions of Real Click Labels

Recent breakthroughs in large models have highlighted the critical significance of data scale, labels and modals. In this paper, we introduce MS MARCO Web Search, the first large-scale information-rich web dataset, featuring millions of real clicked query-document labels. This dataset closely mimics real-world web document and query distribution, provides rich information for various kinds of downstream tasks and encourages research in various areas, such as generic end-to-end neural indexer models, generic embedding models, and next generation information access system with large language models. MS MARCO Web Search offers a retrieval benchmark with three web retrieval challenge tasks that demand innovations in both machine learning and information retrieval system research domains. As the first dataset that meets large, real and rich data requirements, MS MARCO Web Search paves the way for future advancements in AI and system research. MS MARCO Web Search dataset is available at: https://github.com/microsoft/MS-MARCO-Web-Search.

cs.IR↗

arXiv2024

Corporate Communication Companion (CCC): An LLM-empowered Writing Assistant for Workplace Social Media

Workplace social media platforms enable employees to cultivate their professional image and connect with colleagues in a semi-formal environment. While semi-formal corporate communication poses a unique set of challenges, large language models (LLMs) have shown great promise in helping users draft and edit their social media posts. However, LLMs may fail to capture individualized tones and voices in such workplace use cases, as they often generate text using a "one-size-fits-all" approach that can be perceived as generic and bland. In this paper, we present Corporate Communication Companion (CCC), an LLM-empowered interactive system that helps people compose customized and individualized workplace social media posts. Using need-finding interviews to motivate our system design, CCC decomposes the writing process into two core functions, outline and edit: First, it suggests post outlines based on users' job status and previous posts, and next provides edits with attributions that users can contextually customize. We conducted a within-subjects user study asking participants both to write posts and evaluate posts written by others. The results show that CCC enhances users' writing experience, and audience members rate CCC-enhanced posts as higher quality than posts written using a non-customized writing assistant. We conclude by discussing the implications of LLM-empowered corporate communication.

cs.HC↗