17876 research outputs found
Sort by
Think or Step-by-Step? UnZIPping the Black Box in Zero-Shot Prompts
Full text linkZero-shot instructional prompts like “Let’s think step-by-step” have dramatically improved Large Language Model (LLM) performance, yet we lack systematic understanding of which words in these prompts drive their effectiveness. Is think or step-by-step more crucial? How much do seemingly minor word choices affect model outputs? We introduce the ZIP score (Zero-shot Importance of Perturbation score), a metric that quantifies individual word importance through systematic perturbations including synonym replacement, co-hyponym substitution, and word removal. Our experiments reveal individual word choices can have unexpected and counterintuitive effects on model performance. Across four flagship models, seven widely-used prompts, and several tasks, we find that: (1) word importance varies significantly by task type, with mathematical tasks prioritizing structured instruction words like “step-by-step” while reasoning tasks favor “think”; (2) proprietary models align more closely with human judgments than open-source ones; (3) nouns consistently show the highest importance (47.4%–65.9%); and (4) ZIP scores inversely correlate with model performance (|r| > 0.9), indicating stronger word-level effects on challenging tasks. These findings advance prompt science, the systematic study of how prompts elicit model behavior, contributing to both more effective prompt design and improved understanding of word-level effects in LLMs
Is it just a numbers game? An examination of the relationship between exposure to calorie labels and body-related self-conscious emotions
Full text linkMillions of people view calorie information daily across various mediums including on restaurant menus and social media posts [e.g., What I Eat in a Day (WIEIAD) videos]. However, there is a paucity of research on the relationship between viewing calorie information and body image. The purpose of this dissertation was to examine the relationship between viewing calories and negative body image. In study 1 (qualitative study), participants (N = 34) discussed experiences of shame, guilt, and hubristic pride in relation to viewing calories on restaurant menus. In study 2 (experimental design), men and women (N = 359, Mage = 42 years) who viewed a menu with calories reported significantly higher body-related shame, but not body-related guilt or hubristic pride, than those who viewed the same menu without calories. Self-compassion moderated the relationship between exposure to calories on menus and body-related shame; negative effects of viewing calories on menus were strongest for participants low in self-compassion.
In study 3 (experimental design), young women (N = 335, Mage = 25 years) who viewed WIEIAD videos (with or without calories) reported higher body-related envy and intentions to diet than participants who viewed travel videos, with no differences between groups viewing WIEIAD videos with or without calories. Appearance comparisons mediated the relationship between WIEIAD videos and body-related shame, guilt, envy, and intentions to change diet, however self-compassion did not moderate any relationships.
In study 4, young men (N = 333, Mage = 25 years) who viewed WIEIAD videos with calories reported higher fitness-related shame and guilt compared to participants who viewed WIEIAD videos without calories and travel videos. Appearance comparisons mediated the effect of viewing WIEIAD videos (with and without calories) on all outcomes, with higher appearance comparisons leading to greater fitness-related shame, guilt, envy, and intentions to change diet and exercise. Self-compassion did not moderate any effects.
Overall, findings highlight the complexity of the relationship between viewing calories and negative body-related self-conscious emotions, as findings differed based on several factors including how calories were viewed. In addition, self-compassion seemed protective in some, but not all contexts
TAAF: A Trace Abstraction and Analysis Framework Synergizing Knowledge Graphs and LLMs
No full textExecution trace data contains a rich source of information crucial for understanding, debugging, and optimizing software executions. However, traces generated from operating systems or complex applications like Chrome or MySQL are often extremely large and difficult to analyze. Existing trace analysis and visualization tools typically rely on predefined analyses. When users need specific, customized insights, they often find it either impossible with these tools or must develop their own analyses, which is time-consuming and requires significant domain knowledge. Even when suitable predefined analyses exist, users must manually locate the specific analysis, open it, scroll through extensive kernel events to identify particular timestamps, and finally interpret values that often still require expert knowledge to fully comprehend. This research introduces the Trace Abstraction and Analysis Framework (TAAF), a novel approach that integrates large language models (LLMs), knowledge graphs (KGs), and time-indexing techniques to bridge the gap between raw trace data and actionable insights. TAAF constructs a time-indexed knowledge graph from execution trace events, capturing structural and contextual relationships among threads, CPUs, and key system attributes. Generative AI models then utilize these knowledge graphs to answer user queries expressed in natural language, significantly reducing the manual effort and specialized expertise traditionally required in trace analysis. To validate TAAF we present TraceQA-100, a 100-question benchmark built on kernel traces, and run extensive experiments on this suite. Results demonstrate that combining knowledge graphs with generative AI within TAAF improves answer quality and accuracy up to 31.2% compared to manual methods or raw data alone, particularly in tasks involving multi-hop reasoning and causal analysis. We also examine the strengths and limitations of applying LLMs and KGs to trace analysis
Choice, Challenge, and Consequence in Branching Narrative Games
No full textAs interactive media grows in sophistication and complexity, players increasingly seek narratives that respond to their input. However, many game design methods still prioritize mechanics, an interaction that is designed to take place within the game world, over the storytelling elements or narrative. This project introduces a revised four-pillar model built around agency, context, mechanics, and conveyance. These pillars assess three titles: As Dusk Falls, Bandersnatch, and Slay the Princess, examining each for how it handles narrative interaction and whether it supports or limits a player’s agency. The findings inform the creation of a narrative-driven prototype showcasing that even simple systems can support strong narrative design when making the correct considerations to the aforementioned pillars. When the player’s input is made visible through consistent writing and feedback, agency is preserved with or without complex mechanics. The model offers a method for structuring narrative systems that champion player interaction
Towards Personalized LLMs: Investigating Narrative Generation and Personality-Based Preferences in Large Language Models
Full text linkLarge Language Models (LLMs) have transformed text generation, yet aligning their outputs with individual identities and preferences remains underexplored. This research addresses these challenges by (1) evaluating LLMs’ capabilities in generating personalized narratives and their impacts and (2) examining how diverse personality traits shape human-LLM interactions.
In the first part, we explore the effectiveness of LLMs in generating personalized “mirror stories” that reflect and resonate with individual readers’ identities. We introduce MirrorStories, a corpus of 1,500 personalized short stories incorporating elements such as name, gender, age, ethnicity, interests, and moral themes. Our evaluation with 26 diverse human judges shows that LLMs effectively integrate these identity elements, with personalized stories outperforming generic narratives in engagement, satisfaction and personal relevance. We also analyze biases in generated content and the integration of images into personalized storytelling.
The second part investigates whether personality traits influence preferences for different LLMs. In a study with 32 participants evenly split across four Keirsey personality types, users engaged with GPT-4 and Claude 3.5 across four collaborative tasks: data analysis, creative writing, information retrieval, and writing assistance. Findings reveal personality-based preferences: Rationals favored GPT-4 for goal-oriented tasks, while Idealists preferred Claude 3.5 for creative and open-ended ones. Other types showed more task-dependent preferences. Sentiment analysis of participant feedback supported these trends. Interestingly, overall helpfulness ratings were similar across models, showing how personality-based analysis reveals LLM differences that traditional evaluations miss
The Longitudinal Links between Earlier HEXACO and Later Dark Triad traits in Adolescents
Full text linkIt can be difficult to study the Dark Triad (psychopathy, narcissism, Machiavellianism) in adolescents due to concerns over stigma or applying nonspecific or pejorative labels, and one potential solution to this problem is to use broader measures rather than specific Dark ones. Prior research suggests that the best broader measure is the HEXACO personality inventory, given its conceptual and theoretical overlap with the Dark Triad. There is a lack of longitudinal research examining the overlap between HEXACO and the Dark Triad in adolescents. To fill this gap, we conducted a path model analysis in a community sample of adolescents (N = 388) to determine what self-reported HEXACO traits at Time 1 predicted Dark Triad traits at Time 2, six months later. Psychopathy at Time 2 was predicted by lower Honesty-Humility and lower Conscientiousness at Time 1, Machiavellianism was predicted by lower Agreeableness and Emotionality, and higher Conscientiousness at Time 1, and narcissism was predicted by lower Honesty-Humility, Agreeableness, Emotionality, and higher Extraversion at Time 1. What we suggest based on these findings is that there are earlier “clusters” of HEXACO traits that could potentially develop into later Dark Triad traits, which has implications for potential earlier interventions. Further implications for research are discussed
Back Into My Body: An Autoethnography
Full text linkFor the culminating task in my Master of Education (M.Ed) journey, I completed this autoethnographic study, in which I recounted and examined the journey back into my body, reclaiming the parts of myself long dissociated due to a life lived on the margins of identity, culture, and geography. Framed by Mezirow’s Transformative Learning Theory (TLT) (Fleming, 2018; Paprock, 1992; Mezirow, 1991), and Womanism (DeLoach & Young, 2014; Handy, 2002; Maparyan, 2018; Pratt, 2023; Weida, 2024), I examined how mind/body dualism (Damasio, 2006; Morrock, 1973) – entrenched in colonial, neoliberal, white supremacist patriarchy (hooks, 2013) – fractured my sense of self across two members of the British Commonwealth, Jamaica, and Canada. Through vignettes, reflection, and storytelling, I explored how familial, educational, and institutional structures forced me into conformity while rendering my body invisible and my voice inaudible (Adams & Zúñiga, 2016; Maté, 2022; Pheterson, 1986).
Despite academic and professional achievements and social status, I was persistently “othered” (Douglass, 1991; Wayland, 2006), expected to display gratitude while masking pain. My healing began not in textbooks, but through somatic movement, nature, prayer, and integrative energy therapy. Each practice helped me reinhabit my body, reclaim my emotions, and break from the binary thinking that once confined me (Blackwell, 2023; Chawla & Atay, 2018; Medwinter & Rozario, 2021; Miller & Tran, 2024; Smith, 2021).
This work is both my liberation and my offering. I now see my life at the fringes not as marginal, but as multidimensional. My body is no longer a site of suppression but a vessel of wisdom, love, and sovereignty (hooks, 2013; hooks, 2001). As an educator, I hope my story invites others toward wholeness – and into the radical, embodied practice of freedom (hooks, 1994)
Two-Dimensional Titanium Carbide MXene for Lithium-ion Conductive Solid-state Electrolytes
No full textThis study presents the fabrication and comprehensive characterization of polymer-based lithium-ion conductive solid-state electrolytes (SPEs) incorporating two-dimensional (2D) titanium carbide (Ti3C2Tx) MXene as a nanofiller. 2D Ti3C2Tx MXene was synthesized using a mild etching approach involving low-toxicity acids, enabling a safer and more environmentally friendly alternative to traditional synthesis methods. The resulting MXene was characterized using X-ray diffraction (XRD), scanning electron microscopy with energy-dispersive spectroscopy (SEM-EDS), Fourier-transform infrared (FTIR) spectroscopy, atomic force microscopy (AFM), and Kelvin probe force microscopy (KPFM) to evaluate its structural and morphological properties. To explore the potential of 2D Ti3C2Tx MXene in SPEs, the MXene were integrated into three different polymer matrices and the resulting SPEs were evaluated through a suite of advanced characterization techniques. Electrochemical impedance spectroscopy (EIS) was used to study ionic conductivity under varying temperature and humidity conditions. Differential scanning calorimetry (DSC) assessed thermal behavior, and universal testing system (UTS) was used to evaluate mechanical performance of fabricated SPEs. SEM-EDS and FTIR further elucidated the morphological and chemical characteristics of the SPEs. The developed MXene-based SPEs demonstrated substantial improvements in ionic conductivity, as well as enhanced thermal, mechanical, and structural stability, exceeding established benchmarks for SPEs. These results highlight the promising potential of 2D Ti3C2Tx MXene nanofillers in advancing the performance and safety of next-generation solid-state lithium-ion and lithium metal batteries
The effects of parental experience on cognition, anxiety-like behaviour, and hippocampal plasticity in a biparental species.
Full text linkBecoming a parent is a transformative event marked by significant changes in behaviour and the brain. In rodents, it is still unclear how parental experience influences behaviours outside of caregiving, such as in spatial cognition and anxiety. Furthermore, studies on the effects of maternal experience have focused on monoparental species, while few studies have investigated the changes associated with fatherhood. To date, a direct comparison on the effects of parental experience on behaviour and the brain in both mothers and fathers of the same species has not been done. The objective of this study was to investigate the effects of parental experience on spatial cognition, anxiety-like behaviour, and hippocampal neuroplasticity-related measures (microglia and perineuronal net expression) in both sexes of the same species, the degu. Degus are biparental rodents allowing us to examine maternal and paternal experiences in addition to maternal experience in single mothers when the male partner is removed (i.e., monoparental maternal experience). Key findings from our study indicated that parental experience differentially affects anxiety-like behaviour and spatial learning and memory in males and females. Biparental females exhibited more anxiogenic behaviour while biparental males showed more anxiolytic behaviour on the elevated plus maze. Furthermore, biparental males exhibited impaired spatial learning, while monoparental females exhibited enhanced spatial learning on the Barnes maze. In the hippocampus, parental experience did not affect the density of microglia and the expression of perineuronal nets in either the dorsal or ventral dentate gyrus. These results demonstrate that parenthood remodels behaviour and affects anxiety and spatial cognition in a differential manner across sexes. However, these alterations in behaviour do not appear to be associated with changes in microglia or perineuronal net expression in the dentate gyrus, suggesting alternative regions and mechanisms are involved
