arXiv subjects

Kimberly Le Truong

Publications and source records attributed to Kimberly Le Truong.

4 recordsLinked to original sources

arXiv2026

A Framework for Generating Valid Context-Specific Benchmarks through Expert Guidance

This paper presents an end-to-end approach for generating context-specific large language model (LLM) benchmark datasets by combining expert input with synthetic data generation. Existing benchmark construction methods often trade off validity and scalability: datasets designed with domain experts can produce high-quality evaluations but are slow and costly to create, while synthetically generating data may scale efficiently but often results in unrealistic, redundant, or out-of-scope examples. To address this gap, we introduce a schema eliciting key information about the goals, scope, and context of an evaluation task, and use this information to guide synthetic data generation. We further define four criteria grounded in measurement validity for assessing dataset quality: coverage, diversity, content realism, and stylistic realism. Using these criteria, we show how expert-informed scaffolds can guide synthetic data generation toward more valid benchmarks. Through quantitative evaluations and a real-world case study with domain experts, we demonstrate that our approach improves benchmark data quality over existing methods while preserving validity. We additionally analyze how different types of schema information affect different dataset quality criteria, and provide practical guidance on which information to prioritize collecting under resource constraints.

cs.AI↗

arXiv2026

Justified or Just Convincing? Error Verifiability as a Dimension of LLM Quality

As LLMs are deployed in high-stakes settings, users must judge the correctness of individual responses, often relying on model-generated justifications such as reasoning chains or explanations. Yet, no standard measure exists for whether these justifications help users distinguish correct answers from incorrect ones. We formalize this idea as error verifiability and propose $v_{\text{bal}}$, a balanced metric that measures whether justifications enable raters to accurately assess answer correctness, validated against human raters who show high agreement. We find that neither common approaches, such as post-training and model scaling, nor more targeted interventions recommended improve verifiability. We introduce two methods that succeed at improving verifiability: reflect-and-rephrase (RR) for mathematical reasoning and oracle-rephrase (OR) for factual QA, both of which improve verifiability by incorporating domain-appropriate external information. Together, our results establish error verifiability as a distinct dimension of response quality that does not emerge from accuracy improvements alone and requires dedicated, domain-aware methods to address.

cs.HC↗

arXiv2025

Persona-Augmented Benchmarking: Evaluating LLMs Across Diverse Writing Styles

Current benchmarks for evaluating Large Language Models (LLMs) often do not exhibit enough writing style diversity, with many adhering primarily to standardized conventions. Such benchmarks do not fully capture the rich variety of communication patterns exhibited by humans. Thus, it is possible that LLMs, which are optimized on these benchmarks, may demonstrate brittle performance when faced with "non-standard" input. In this work, we test this hypothesis by rewriting evaluation prompts using persona-based LLM prompting, a low-cost method to emulate diverse writing styles. Our results show that, even with identical semantic content, variations in writing style and prompt formatting significantly impact the estimated performance of the LLM under evaluation. Notably, we identify distinct writing styles that consistently trigger either low or high performance across a range of models and tasks, irrespective of model family, size, and recency. Our work offers a scalable approach to augment existing benchmarks, improving the external validity of the assessments they provide for measuring LLM performance across linguistic variations.

cs.CL↗

arXiv2025

Toward Valid Measurement Of (Un)fairness For Generative AI: A Proposal For Systematization Through The Lens Of Fair Equality of Chances

Disparities in the societal harms and impacts of Generative AI (GenAI) systems highlight the critical need for effective unfairness measurement approaches. While numerous benchmarks exist, designing valid measurements requires proper systematization of the unfairness construct. Yet this process is often neglected, resulting in metrics that may mischaracterize unfairness by overlooking contextual nuances, thereby compromising the validity of the resulting measurements. Building on established (un)fairness measurement frameworks for predictive AI, this paper focuses on assessing and improving the validity of the measurement task. By extending existing conceptual work in political philosophy, we propose a novel framework for evaluating GenAI unfairness measurement through the lens of the Fair Equality of Chances framework. Our framework decomposes unfairness into three core constituents: the harm/benefit resulting from the system outcomes, morally arbitrary factors that should not lead to inequality in the distribution of harm/benefit, and the morally decisive factors, which distinguish subsets that can justifiably receive different treatments. By examining fairness through this structured lens, we integrate diverse notions of (un)fairness while accounting for the contextual dynamics that shape GenAI outcomes. We analyze factors contributing to each component and the appropriate processes to systematize and measure each in turn. This work establishes a foundation for developing more valid (un)fairness measurements for GenAI systems.

cs.CY↗