arXiv subjects

Kristina Gligorić

Publications and source records attributed to Kristina Gligorić.

At least 19 recordsLinked to original sources

arXiv2026

Cooking Up Risks: Benchmarking and Reducing Food Safety Risks in Large Language Models

Food safety directly affects human health in daily life, as unsafe preparation or storage can lead to contamination and foodborne illness. PwC's 2025 consumer survey reports that 47% of respondents are comfortable using generative AI for meal planning, highlighting the need for safe food-related assistance. Yet the safety of large language models (LLMs) against malicious food-related requests remains underexplored. We introduce FoodGuardBench, an FDA-grounded benchmark with 2,124 seed malicious questions, 16,992 variants from eight jailbreak strategies, and 1,000 benign questions across five food-safety risk categories. Across twelve LLMs, the mean overall ASR is 25.51%, whereas the average per-model maximum across eight attacks is 60.59%, revealing that testing only one attack can substantially underestimate the full extent of food safety risk. For several models, nearly half or more of unsafe answers are rated as ready-to-use guides for harmful actions. Existing general-purpose guardrails average 96.25% benign utility but only 55.80% detection accuracy against these eight jailbreak attacks. Using FoodGuardBench, we fine-tune FoodGuard-4B, a food-safety-specific guardrail that achieves 97.64% out-of-distribution (OOD) accuracy while maintaining high benign utility. This demonstrates the value of FoodGuardBench for identifying and mitigating food safety risks.

cs.CR↗

arXiv2026

Multi-Perspective LLM Annotations for Valid Analyses in Subjective Tasks

Large language models are increasingly used to annotate texts, but their outputs reflect some human perspectives better than others. Existing methods for correcting LLM annotation error assume a single ground truth. However, this assumption fails in subjective tasks where disagreement across demographic groups is meaningful. Here we introduce Perspective-Driven Inference, a method that treats the distribution of annotations across groups as the quantity of interest, and estimates it using a small human annotation budget. We contribute an adaptive sampling strategy that concentrates human annotation effort on groups where LLM proxies are least accurate. We evaluate on politeness and offensiveness rating tasks, showing targeted improvements for harder-to-model demographic groups relative to uniform sampling baselines, while maintaining coverage.

cs.CL↗

arXiv2026

SWAY: A Counterfactual Computational Linguistic Approach to Measuring and Mitigating Sycophancy

Large language models exhibit sycophancy: the tendency to shift outputs toward user-expressed stances, regardless of correctness or consistency. While prior work has studied this issue and its impacts, rigorous computational linguistic metrics are needed to identify when models are being sycophantic. Here, we introduce SWAY, an unsupervised computational linguistic measure of sycophancy. We develop a counterfactual prompting mechanism to identify how much a model's agreement shifts under positive versus negative linguistic pressure, isolating framing effects from content. Applying this metric to benchmark 6 models, we find that sycophancy increases with epistemic commitment. Leveraging our metric, we introduce a counterfactual mitigation strategy teaching models to consider what the answer would be if opposite assumptions were suggested. While baseline mitigation instructing to be explicitly anti-sycophantic yields moderate reductions, and can backfire, our counterfactual CoT mitigation drives sycophancy to near zero across models, commitment levels, and clause types, while not suppressing responsiveness to genuine evidence. Overall, we contribute a metric for benchmarking sycophancy and a mitigation informed by it.

cs.CL↗

arXiv2025

Valid Survey Simulations with Limited Human Data: The Roles of Prompting, Fine-Tuning, and Rectification

Surveys provide valuable insights into public opinion and behavior, but their execution is costly and slow. Large language models (LLMs) have been proposed as a scalable, low-cost substitute for human respondents, but their outputs are often biased and yield invalid estimates. We study the interplay between synthesis methods that use LLMs to generate survey responses and rectification methods that debias population estimates, and explore how human responses are best allocated between them. Using two panel surveys with questions on nutrition, politics, and economics, we find that synthesis alone introduces substantial bias (24-86%), whereas combining it with rectification reduces bias below 5% and increases effective sample size by up to 14%. Overall, we challenge the common practice of using all human responses for fine-tuning, showing that under a fixed budget, allocating most to rectification results in far more effective estimation.

cs.CL↗

arXiv2025

Attention to Non-Adopters

Although language model-based chat systems are increasingly used in daily life, most Americans remain non-adopters of chat-based LLMs -- as of June 2025, 66% had never used ChatGPT. At the same time, LLM development and evaluation rely mainly on data from adopters (e.g., logs, preference data), focusing on the needs and tasks for a limited demographic group of adopters in terms of geographic location, education, and gender. In this position paper, we argue that incorporating non-adopter perspectives is essential for developing broadly useful and capable LLMs. We contend that relying on methods that focus primarily on adopters will risk missing a range of tasks and needs prioritized by non-adopters, entrenching inequalities in who benefits from LLMs, and creating oversights in model development and evaluation. To illustrate this claim, we conduct case studies with non-adopters and show: how non-adopter needs diverge from those of current users, how non-adopter needs point us towards novel reasoning tasks, and how to systematically integrate non-adopter needs via human-centered methods.

cs.CY↗

arXiv2025

What can large language models do for sustainable food?

Food systems are responsible for a third of human-caused greenhouse gas emissions. We investigate what Large Language Models (LLMs) can contribute to reducing the environmental impacts of food production. We define a typology of design and prediction tasks based on the sustainable food literature and collaboration with domain experts, and evaluate six LLMs on four tasks in our typology. For example, for a sustainable protein design task, food science experts estimated that collaboration with an LLM can reduce time spent by 45% on average, compared to 22% for collaboration with another expert human food scientist. However, for a sustainable menu design task, LLMs produce suboptimal solutions when instructed to consider both human satisfaction and climate impacts. We propose a general framework for integrating LLMs with combinatorial optimization to improve reasoning capabilities. Our approach decreases emissions of food choices by 79% in a hypothetical restaurant while maintaining participants' satisfaction with their set of choices. Our results demonstrate LLMs' potential, supported by optimization techniques, to accelerate sustainable food development and adoption.

cs.CY↗

arXiv2025

Meat-Free Day Reduces Greenhouse Gas Emissions but Poses Challenges for Customer Retention and Adherence to Dietary Guidelines

Reducing meat consumption is crucial for achieving global environmental and nutritional targets. Meat-Free Day (MFD) is a widely adopted strategy to address this challenge by encouraging plant-based diets through the removal of animal-based meals. We assessed the environmental, behavioral, and nutritional impacts of MFD by implementing 67 MFDs over 18 months (once a week on a randomly chosen day) across 12 cafeterias on a large university campus, analyzing over 400,000 food purchases. MFD reduced on-campus food-related greenhouse gas (GHG) emissions on treated days by 52.9% and contributed to improved fiber (+26.9%) and cholesterol (-4.5%) consumption without altering caloric intake. These nutritional benefits were, however, accompanied by a 27.6% decrease in protein intake and a 34.2% increase in sugar consumption. Moreover, the increase in plant-based meals did not carry over to subsequent days, as evidenced by a 3.5% rebound in animal-based meal consumption on days immediately following treated days. MFD also led to a 16.8% drop in on-campus meal sales on treated days.Monte Carlo simulations suggest that if 8.7% of diners were to eat burgers off-campus on treated days, MFD's GHG savings would be fully negated. As our analysis identifies on-campus customer retention as the main challenge to MFD effectiveness, we recommend combining MFD with customer retention interventions to ensure environmental and nutritional benefits.

cs.CY↗

arXiv2025

Can Unconfident LLM Annotations Be Used for Confident Conclusions?

Large language models (LLMs) have shown high agreement with human raters across a variety of tasks, demonstrating potential to ease the challenges of human data collection. In computational social science (CSS), researchers are increasingly leveraging LLM annotations to complement slow and expensive human annotations. Still, guidelines for collecting and using LLM annotations, without compromising the validity of downstream conclusions, remain limited. We introduce Confidence-Driven Inference: a method that combines LLM annotations and LLM confidence indicators to strategically select which human annotations should be collected, with the goal of producing accurate statistical estimates and provably valid confidence intervals while reducing the number of human annotations needed. Our approach comes with safeguards against LLM annotations of poor quality, guaranteeing that the conclusions will be both valid and no less accurate than if we only relied on human annotations. We demonstrate the effectiveness of Confidence-Driven Inference over baselines in statistical estimation tasks across three CSS settings--text politeness, stance, and bias--reducing the needed number of human annotations by over 25% in each. Although we use CSS settings for demonstration, Confidence-Driven Inference can be used to estimate most standard quantities across a broad range of NLP problems.

cs.CL↗

arXiv2024

Grounding Gaps in Language Model Generations

Effective conversation requires common ground: a shared understanding between the participants. Common ground, however, does not emerge spontaneously in conversation. Speakers and listeners work together to both identify and construct a shared basis while avoiding misunderstanding. To accomplish grounding, humans rely on a range of dialogue acts, like clarification (What do you mean?) and acknowledgment (I understand.). However, it is unclear whether large language models (LLMs) generate text that reflects human grounding. To this end, we curate a set of grounding acts and propose corresponding metrics that quantify attempted grounding. We study whether LLM generations contain grounding acts, simulating turn-taking from several dialogue datasets and comparing results to humans. We find that -- compared to humans -- LLMs generate language with less conversational grounding, instead generating text that appears to simply presume common ground. To understand the roots of the identified grounding gap, we examine the role of instruction tuning and preference optimization, finding that training on contemporary preference data leads to a reduction in generated grounding acts. Altogether, we highlight the need for more research investigating conversational grounding in human-AI interaction.

cs.CL↗

arXiv2024

Othering and low status framing of immigrant cuisines in US restaurant reviews and large language models

Identifying implicit attitudes toward food can mitigate social prejudice due to food's salience as a marker of ethnic identity. Stereotypes about food are representational harms that may contribute to racialized discourse and negatively impact economic outcomes for restaurants. Understanding the presence of representational harms in online corpora in particular is important, given the increasing use of large language models (LLMs) for text generation and their tendency to reproduce attitudes in their training data. Through careful linguistic analyses, we evaluate social theories about attitudes toward immigrant cuisine in a large-scale study of framing differences in 2.1M English language Yelp reviews. Controlling for factors such as restaurant price and neighborhood racial diversity, we find that immigrant cuisines are more likely to be othered using socially constructed frames of authenticity (e.g., "authentic," "traditional"), and that non-European cuisines (e.g., Indian, Mexican) in particular are described as more exotic compared to European ones (e.g., French). We also find that non-European cuisines are more likely to be described as cheap and dirty, even after controlling for price, and even among the most expensive restaurants. Finally, we show that reviews generated by LLMs reproduce similar framing tendencies, pointing to the downstream retention of these representational harms. Our results corroborate social theories of gastronomic stereotyping, revealing racialized evaluative processes and linguistic strategies through which they manifest.

cs.CL↗

arXiv2022

Anticipated versus Actual Effects of Platform Design Change: A Case Study of Twitter's Character Limit

The design of online platforms is both critically important and challenging, as any changes may lead to unintended consequences, and it can be hard to predict how users will react. Here we conduct a case study of a particularly important real-world platform design change: Twitter's decision to double the character limit from 140 to 280 characters to soothe users' need to ''cram'' or ''squeeze'' their tweets, informed by modeling of historical user behavior. In our analysis, we contrast Twitter's anticipated pre-intervention predictions about user behavior with actual post-intervention user behavior: Did the platform design change lead to the intended user behavior shifts, or did a gap between anticipated and actual behavior emerge? Did different user groups react differently? We find that even though users do not ''cram'' as much under 280 characters as they used to under 140 characters, emergent ``cramming'' at the new limit seems to not have been taken into account when designing the platform change. Furthermore, investigating textual features, we find that, although post-intervention ''crammed'' tweets are longer, their syntactic and semantic characteristics remain similar and indicative of ''squeezing''. Applying the same approach as Twitter policy-makers, we create updated counterfactual estimates and find that the character limit would need to be increased further to reduce cramming that re-emerged at the new limit. We contribute to the rich literature studying online user behavior with an empirical study that reveals a dynamic interaction between platform design and user behavior, with immediate policy and practical implications for the design of socio-technical systems.

cs.SI↗

arXiv2022

Biased Bytes: On the Validity of Estimating Food Consumption from Digital Traces

Given that measuring food consumption at a population scale is a challenging task, researchers have begun to explore digital traces (e.g., from social media or from food-tracking applications) as potential proxies. However, it remains unclear to what extent digital traces reflect real food consumption. The present study aims to bridge this gap by quantifying the link between dietary behaviors as captured via social media (Twitter) v.s. a food-tracking application (MyFoodRepo). We focus on the case of Switzerland and contrast images of foods collected through the two platforms, by designing and deploying a novel crowdsourcing framework for estimating biases with respect to nutritional properties and appearance. We find that the food type distributions in social media v.s. food tracking diverge; e.g., bread is 2.5 times more frequent among consumed and tracked foods than on Twitter, whereas cake is 12 times more frequent on Twitter. Controlling for the different food type distributions, we contrast consumed and tracked foods of a given type with foods shared on Twitter. Across food types, food posted on Twitter is perceived as tastier, more caloric, less healthy, less likely to have been consumed at home, more complex, and larger-portioned, compared to consumed and tracked foods. The fact that there is a divergence between food consumption as measured via the two platforms implies that at least one of the two is not a faithful representation of the true food consumption in the general Swiss population. Thus, researchers should be attentive and aim to establish evidence of validity before using digital traces as a proxy for the true food consumption of a general population. We conclude by discussing the potential sources of these biases and their implications, outlining pitfalls and threats to validity, and proposing actionable ways for overcoming them.

cs.SI↗

arXiv2021

Laughing Heads: Can Transformers Detect What Makes a Sentence Funny?

The automatic detection of humor poses a grand challenge for natural language processing. Transformer-based systems have recently achieved remarkable results on this task, but they usually (1)~were evaluated in setups where serious vs humorous texts came from entirely different sources, and (2)~focused on benchmarking performance without providing insights into how the models work. We make progress in both respects by training and analyzing transformer-based humor recognition models on a recently introduced dataset consisting of minimal pairs of aligned sentences, one serious, the other humorous. We find that, although our aligned dataset is much harder than previous datasets, transformer-based models recognize the humorous sentence in an aligned pair with high accuracy (78%). In a careful error analysis, we characterize easy vs hard instances. Finally, by analyzing attention weights, we obtain important insights into the mechanisms by which transformers recognize humor. Most remarkably, we find clear evidence that one single attention head learns to recognize the words that make a test sentence humorous, even without access to this information at training time.

cs.CL↗

arXiv2021

Sudden Attention Shifts on Wikipedia During the COVID-19 Crisis

We study how the COVID-19 pandemic, alongside the severe mobility restrictions that ensued, has impacted information access on Wikipedia, the world's largest online encyclopedia. A longitudinal analysis that combines pageview statistics for 12 Wikipedia language editions with mobility reports published by Apple and Google reveals massive shifts in the volume and nature of information seeking patterns during the pandemic. Interestingly, while we observe a transient increase in Wikipedia's pageview volume following mobility restrictions, the nature of information sought was impacted more permanently. These changes are most pronounced for language editions associated with countries where the most severe mobility restrictions were implemented. We also find that articles belonging to different topics behaved differently; e.g., attention towards entertainment-related topics is lingering and even increasing, while the interest in health- and biology-related topics was either small or transient. Our results highlight the utility of Wikipedia for studying how the pandemic is affecting people's needs, interests, and concerns.

cs.CY↗

arXiv2021

Formation of Social Ties Influences Food Choice: A Campus-Wide Longitudinal Study

Nutrition is a key determinant of long-term health, and social influence has long been theorized to be a key determinant of nutrition. It has been difficult to quantify the postulated role of social influence on nutrition using traditional methods such as surveys, due to the typically small scale and short duration of studies. To overcome these limitations, we leverage a novel source of data: logs of 38 million food purchases made over an 8-year period on the Ecole Polytechnique Federale de Lausanne (EPFL) university campus, linked to anonymized individuals via the smartcards used to make on-campus purchases. In a longitudinal observational study, we ask: How is a person's food choice affected by eating with someone else whose own food choice is healthy vs. unhealthy? To estimate causal effects from the passively observed log data, we control confounds in a matched quasi-experimental design: we identify focal users who at first do not have any regular eating partners but then start eating with a fixed partner regularly, and we match focal users into comparison pairs such that paired users are nearly identical with respect to covariates measured before acquiring the partner, where the two focal users' new eating partners diverge in the healthiness of their respective food choice. A difference-in-differences analysis of the paired data yields clear evidence of social influence: focal users acquiring a healthy-eating partner change their habits significantly more toward healthy foods than focal users acquiring an unhealthy-eating partner. We further identify foods whose purchase frequency is impacted significantly by the eating partner's healthiness of food choice. Beyond the main results, the work demonstrates the utility of passively sensed food purchase logs for deriving insights, with the potential of informing the design of public health interventions and food offerings.

cs.SI↗

arXiv2020

Adoption of Twitter's New Length Limit: Is 280 the New 140?

In November 2017, Twitter doubled the maximum allowed tweet length from 140 to 280 characters, a drastic switch on one of the world's most influential social media platforms. In the first long-term study of how the new length limit was adopted by Twitter users, we ask: Does the effect of the new length limit resemble that of the old one? Or did the doubling of the limit fundamentally change how Twitter is shaped by the limited length of posted content? By analyzing Twitter's publicly available 1% sample over a period of around 3 years, we find that, when the length limit was raised from 140 to 280 characters, the prevalence of tweets around 140 characters dropped immediately, while the prevalence of tweets around 280 characters rose steadily for about 6 months. Despite this rise, tweets approaching the length limit have been far less frequent after than before the switch. We find widely different adoption rates across languages and client-device types. The prevalence of tweets around 140 characters before the switch in a given language is strongly correlated with the prevalence of tweets around 280 characters after the switch in the same language, and very long tweets are vastly more popular on Web clients than on mobile clients. Moreover, tweets of around 280 characters after the switch are syntactically and semantically similar to tweets of around 140 characters before the switch, manifesting patterns of message squeezing in both cases. Taken together, these findings suggest that the new 280-character limit constitutes a new, less intrusive version of the old 140-character limit. The length limit remains an important factor that should be considered in all studies using Twitter data.

cs.SI↗

arXiv2020

Experts and authorities receive disproportionate attention on Twitter during the COVID-19 crisis

Timely access to accurate information is crucial during the COVID-19 pandemic. Prompted by key stakeholders' cautioning against an "infodemic", we study information sharing on Twitter from January through May 2020. We observe an overall surge in the volume of general as well as COVID-19-related tweets around peak lockdown in March/April 2020. With respect to engagement (retweets and likes), accounts related to healthcare, science, government and politics received by far the largest boosts, whereas accounts related to religion and sports saw a relative decrease in engagement. While the threat of an "infodemic" remains, our results show that social media also provide a platform for experts and public authorities to be widely heard during a global crisis.

cs.SI↗

arXiv2019

Message Distortion in Information Cascades

Information diffusion is usually modeled as a process in which immutable pieces of information propagate over a network. In reality, however, messages are not immutable, but may be morphed with every step, potentially entailing large cumulative distortions. This process may lead to misinformation even in the absence of malevolent actors, and understanding it is crucial for modeling and improving online information systems. Here, we perform a controlled, crowdsourced experiment in which we simulate the propagation of information from medical research papers. Starting from the original abstracts, crowd workers iteratively shorten previously produced summaries to increasingly smaller lengths. We also collect control summaries where the original abstract is compressed directly to the final target length. Comparing cascades to controls allows us to separate the effect of the length constraint from that of accumulated distortion. Via careful manual coding, we annotate lexical and semantic units in the medical abstracts and track them along cascades. We find that iterative summarization has a negative impact due to the accumulation of error, but that high-quality intermediate summaries result in less distorted messages than in the control case. Different types of information behave differently; in particular, the conclusion of a medical abstract (i.e., its key message) is distorted most. Finally, we compare abstractive with extractive summaries, finding that the latter are less prone to semantic distortion. Overall, this work is a first step in studying information cascades without the assumption that disseminated content is immutable, with implications on our understanding of the role of word-of-mouth effects on the misreporting of science.

cs.SI↗