arXiv subjects

Stephanie Eckman

Publications and source records attributed to Stephanie Eckman.

10 recordsLinked to original sources

arXiv2026

DRIFTLENS: Measuring Memory-Induced Reasoning Drift in Personalized Language Models

Personalization changes what a model says to a user; we show that it can also change the reasoning trajectory used to justify the response. Modern LLMs personalize interactions by storing user attributes, preferences, and prior context, then injecting this information into future prompts. We study whether such memory reshapes reasoning on open-ended questions where no single ground-truth answer exists. To quantify this effect, we introduce DRIFTLENS, a ground-truth-free framework that maps each expressed reasoning step to a value category and measures divergence between a question's no-memory trajectory and its trajectory under injected user-attribute memory. We first validate that DRIFTLENS distinguishes content-free pragmatic noise from substantive reasoning changes. Across four LLMs and 10 user-attribute categories, including age, occupation, and disability, user-attribute memory induces medium-to-large reasoning drift above each model's pragmatic-noise floor, even when final answers remain fluent, on-topic, and plausible. We then evaluate GRPO- and DPO-based post-training methods for reducing drift. Both reduce drift, but neither uniformly dominates; effects on downstream capability, helpfulness, and instruction following are model-and reward-dependent. These results suggest that memory-induced reasoning drift is a measurable and only partly mitigated failure mode of personalized language models.

cs.AI↗

arXiv2026

From Ground Truth to Measurement: A Statistical Framework for Human Labeling

Supervised machine learning assumes that labeled data provide accurate measurements of the concepts models are meant to learn. Yet in practice, human labeling introduces systematic variation arising from ambiguous items, divergent interpretations, and simple mistakes. Machine learning research commonly treats all disagreement as noise, which obscures these distinctions and limits our understanding of what models actually learn. This paper reframes annotation as a measurement process and introduces a statistical framework for decomposing labeling outcomes into interpretable sources of variation: instance difficulty, annotator bias, situational noise, and relational alignment. The framework extends classical measurement-error models to accommodate both shared and individualized notions of truth, reflecting traditional and human label variation interpretations of error, and provides a diagnostic for assessing which regime better characterizes a given task. Applying the proposed model to a multi-annotator natural language inference dataset, we find empirical evidence for all four theorized components and demonstrate the effectiveness of our approach. We conclude with implications for data-centric machine learning and outline how this approach can guide the development of a more systematic science of labeling.

stat.ME↗

arXiv2025

SATA-BENCH: Select All That Apply Benchmark for Multiple Choice Questions

Large language models (LLMs) are increasingly evaluated on single-answer multiple-choice tasks, yet many real-world problems require identifying all correct answers from a set of options. This capability remains underexplored. We introduce SATA-BENCH, the first dedicated benchmark for evaluating LLMs on Select All That Apply (SATA) questions across diverse domains, including reading comprehension, law, and biomedicine. Our evaluation of 27 open-source and proprietary models reveals a significant gap: even the strongest model achieves only 41.8% exact match, exposing LLMs' inability to reliably identify all correct answers. We find that this weakness stems from two core challenges: selection bias - models favor certain choices regardless of content, and count bias - models fail to predict the correct number of answers. To address these issues, we propose Choice Funnel, a decoding strategy that combines token debiasing with adaptive thresholding to guide models toward complete and accurate selections. Choice Funnel achieves up to 29% higher exact match than competitive baselines while reducing inference cost by over 64%. Our findings expose fundamental limitations in current LLMs and introduce a new framework for diagnosing and improving multi-answer reasoning. We release SATA-BENCH and Choice Funnel to promote LLM development for robust decision-making in realistic, multi-answer applications.

cs.CL↗

arXiv2025

The Personalization Trap: How User Memory Alters Emotional Reasoning in LLMs

When an AI assistant remembers that Sarah is a single mother working two jobs, does it interpret her stress differently than if she were a wealthy executive? As personalized AI systems increasingly incorporate long-term user memory, understanding how this memory shapes emotional reasoning is critical. We investigate how user memory affects emotional intelligence in large language models (LLMs) by evaluating 15 models on human-validated emotional intelligence tests. We find that identical scenarios paired with different user profiles produce systematically divergent emotional interpretations. Across validated user-independent emotional scenarios and diverse user profiles, systematic biases emerged in several high-performing LLMs where advantaged profiles received more accurate emotional interpretations. Moreover, LLMs demonstrate significant disparities across demographic factors in emotion reasoning and supportive recommendations tasks, indicating that personalization mechanisms can embed social hierarchies into models' emotional reasoning. These results highlight a key challenge for memory-enhanced AI: systems designed for personalization may reinforce social inequalities. To mitigate these disparities, we curate a general-purpose preference dataset designed to reduce demographic profiles' influence on emotional understanding.

cs.AI↗

arXiv2025

Bias in the Loop: How Humans Evaluate AI-Generated Suggestions

Human-AI collaboration increasingly drives decision-making across industries, from medical diagnosis to content moderation. While AI systems promise efficiency gains by providing automated suggestions for human review, these workflows can trigger cognitive biases that degrade performance. We know little about the psychological factors that determine when these collaborations succeed or fail. We conducted a randomized experiment with 2,784 participants to examine how task design and individual characteristics shape human responses to AI-generated suggestions. Using a controlled annotation task, we manipulated three factors: AI suggestion quality in the first three instances, task burden through required corrections, and performance-based financial incentives. We collected demographics, attitudes toward AI, and behavioral data to assess four performance metrics: accuracy, correction activity, overcorrection, and undercorrection. Two patterns emerged that challenge conventional assumptions about human-AI collaboration. First, requiring corrections for flagged AI errors reduced engagement and increased the tendency to accept incorrect suggestions, demonstrating how cognitive shortcuts influence collaborative outcomes. Second, individual attitudes toward AI emerged as the strongest predictor of performance, surpassing demographic factors. Participants skeptical of AI detected errors more reliably and achieved higher accuracy, while those favorable toward automation exhibited dangerous overreliance on algorithmic suggestions. The findings reveal that successful human-AI collaboration depends not only on algorithmic performance but also on who reviews AI outputs and how review processes are structured. Effective human-AI collaborations require consideration of human psychology: selecting diverse evaluator samples, measuring attitudes, and designing workflows that counteract cognitive biases.

cs.HC↗

arXiv2025

Aligning NLP Models with Target Population Perspectives using PAIR: Population-Aligned Instance Replication

Models trained on crowdsourced annotations may not reflect population views, if those who work as annotators do not represent the broader population. In this paper, we propose PAIR: Population-Aligned Instance Replication, a post-processing method that adjusts training data to better reflect target population characteristics without collecting additional annotations. Using simulation studies on offensive language and hate speech detection with varying annotator compositions, we show that non-representative pools degrade model calibration while leaving accuracy largely unchanged. PAIR corrects these calibration problems by replicating annotations from underrepresented annotator groups to match population proportions. We conclude with recommendations for improving the representativity of training data and model performance.

stat.ME↗

arXiv2025

Mitigating Selection Bias with Node Pruning and Auxiliary Options

Large language models (LLMs) often exhibit systematic preferences for certain answer choices when responding to multiple-choice questions-a behavior known as selection bias. This bias reduces the accuracy and reliability of LLM outputs, limiting their usefulness in decision-critical applications. While prior work has focused on adjusting model inputs or outputs to mitigate this issue, our work takes a fundamentally different approach by identifying and removing the internal sources of bias. We introduce two methods: Bias Node Pruning (BNP), which prunes parameters that contribute to selection bias, and Auxiliary Option Injection (AOI), which introduces an additional answer choice to reduce bias in both white-box and black-box settings. To address the shortcomings of existing evaluation metrics, we propose Choice Kullback-Leibler Divergence (CKLD), a new metric that captures distributional imbalances in model predictions. Experiments on three LLMs across multiple datasets demonstrate that our methods consistently improve answer accuracy while reducing selection bias, providing a robust solution for both open- and closed-source models.

cs.AI↗

arXiv2024

Position: Insights from Survey Methodology can Improve Training Data

Whether future AI models are fair, trustworthy, and aligned with the public's interests rests in part on our ability to collect accurate data about what we want the models to do. However, collecting high-quality data is difficult, and few AI/ML researchers are trained in data collection methods. Recent research in data-centric AI has show that higher quality training data leads to better performing models, making this the right moment to introduce AI/ML researchers to the field of survey methodology, the science of data collection. We summarize insights from the survey methodology literature and discuss how they can improve the quality of training and feedback data. We also suggest collaborative research ideas into how biases in data collection can be mitigated, making models more accurate and human-centric.

cs.HC↗

arXiv2024

Annotation Sensitivity: Training Data Collection Methods Affect Model Performance

When training data are collected from human annotators, the design of the annotation instrument, the instructions given to annotators, the characteristics of the annotators, and their interactions can impact training data. This study demonstrates that design choices made when creating an annotation instrument also impact the models trained on the resulting annotations. We introduce the term annotation sensitivity to refer to the impact of annotation data collection methods on the annotations themselves and on downstream model performance and predictions. We collect annotations of hate speech and offensive language in five experimental conditions of an annotation instrument, randomly assigning annotators to conditions. We then fine-tune BERT models on each of the five resulting datasets and evaluate model performance on a holdout portion of each condition. We find considerable differences between the conditions for 1) the share of hate speech/offensive language annotations, 2) model performance, 3) model predictions, and 4) model learning curves. Our results emphasize the crucial role played by the annotation instrument which has received little attention in the machine learning literature. We call for additional research into how and why the instrument impacts the annotations to inform the development of best practices in instrument design.

stat.ML↗

arXiv2015

Evaluating the quality of survey and administrative data with generalized multitrait-multimethod models

Administrative register data are increasingly important in statistics, but, like other types of data, may contain measurement errors. To prevent such errors from invalidating analyses of scientific interest, it is therefore essential to estimate the extent of measurement errors in administrative data. Currently, however, most approaches to evaluate such errors involve either prohibitively expensive audits or comparison with a survey that is assumed perfect. We introduce the "generalized multitrait-multimethod" (GMTMM) model, which can be seen as a general framework for evaluating the quality of administrative and survey data simultaneously. This framework allows both survey and register to contain random and systematic measurement errors. Moreover, it accommodates common features of administrative data such as discreteness, nonlinearity, and nonnormality, improving similar existing models. The use of the GMTMM model is demonstrated by application to linked survey-register data from the German Federal Employment Agency on income from and duration of employment, and a simulation study evaluates the estimates obtained. KEY WORDS: Measurement error, Latent Variable Models, Official statistics, Register data, Reliability

stat.AP↗