15064 research outputs found
Sort by
From Tool to Companion: An Unsupervised Typology of Human-LLM Interaction
No full textLarge Language Models (LLMs) have rapidly adapted from specialized tools to everyday conversational partners. Rather than following the technical tradition of asking what LLMs can do, this study explores empirically how people actively frame their encounters with them. I drew 440k English-language first-turn prompts from the open-access WildChat corpus on human-ChatGPT interaction, embedded each into a 384-dimensional semantic space (MiniLM-L6-v2), clustered the landscape with unsupervised learning techniques, and interpreted 60 different topic clusters through qualitative close-reading. The process visually typologized the prompts into a “graded spectrum” — instead of categorizing discrete “use cases,” I depicted an interpretable continuum from highly instrumental queries to deeply expressive or esoteric engagements. Another critical methodological finding is that apparent topical “coherence” often reflects authorship concentration rather than genuine consensus — a small cadre of “power-users” contributed thematically narrow yet voluminous exchanges, inflating the semantic density in repeated or industrialized use. This outsized footprint can skew the usage patterns in open-source chatlogs datasets, leading analysts to underestimate the genuine breadth and interweaving of topics while overestimating the prevalence of anthropomorphism and niche role-playing. The findings complicate the simplistic scope that juxtaposes utilitarian and anthropomorphic use of AI. I showed that many users approach ChatGPT not as a neutral tool but as a mutable social actor, strategically anthropomorphized inside local narrative frames. There is a shifting ecology of motives that is prominent in the dataset yet less discussed in existing studies — from affective self-therapy to workplace automation, and with the hidden labour (“playbor”) of heavy users who are quietly shaping public training data and, by extension, future model behavior. The findings foreground the social and emotional stakes of LLM use, and invite designers to treat usage data with caution while honoring the plurality of ways that humans want to converse with — rather than merely through — their machines
Can AI weather models predict out-of-distribution gray swan tropical cyclones?
Full text linkPredicting gray swan weather extremes, which are possible but so rare that they are absent from the training dataset, is a major concern for AI weather models and long-term climate emulators. An important open question is whether AI models can extrapolate from weaker weather events present in the training set to stronger, unseen weather extremes. To test this, we train independent versions of the AI weather model FourCastNet on the 1979–2015 ERA5 dataset with all data, or with Category 3–5 tropical cyclones (TCs) removed, either globally or only over the North Atlantic or Western Pacific basin. We then test these versions of FourCastNet on 2018–2023 Category 5 TCs (gray swans). All versions yield similar accuracy for global weather, but the one trained without Category 3–5 TCs cannot accurately forecast Category 5 TCs, indicating that these models cannot extrapolate from weaker storms. The versions trained without Category 3–5 TCs in one basin show some skill forecasting Category 5 TCs in that basin, suggesting that FourCastNet can generalize across tropical basins. This is encouraging and surprising because regional information is implicitly encoded in inputs. Given that current state-of-the-art AI weather and climate models have similar learning strategies, we expect our findings to apply to other models. Other types of weather extremes need to be similarly investigated. Our work demonstrates that novel learning strategies are needed for AI models to reliably provide early warning or estimated statistics for the rarest, most impactful TCs, and, possibly, other weather extremes
Lawinengefahr: The Impact of Mountains on Vorarlberg’s National Identity
No full textThis thesis argues that the relationship between Vorarlberg, a small, mountainous, and oft-ignored province in western Austria, and its geography has been formative in developing its cultural and economic identity within Austria. Since the eastern border of the province is formed by the Arlberg massif, travel to and from the rest of the Habsburg empire was incredibly difficult. Dangerous mountain conditions inspired folk legends, and people feared traveling through the mountains even when necessary until a tunnel could be built through the rock in 1884. Because of this geographic isolation, Vorarlberg developed differently from the rest of Austria. Its landscape meant that there was not enough land for large-scale agriculture and industry in the valleys, let alone in the small mountain villages. Although farmers did manage to raise a small number of dairy cattle and grow crops (creating the famous Bergkäse the region is still known for), mountain dwellers struggled to grow and thrive. Luckily, the Romantic movement and improved train infrastructure brought tourists who were in search of the sublime and a mountain adventure. These tourists needed food and lodging, which Vorarlbergers were happy to provide. This nascent tourism industry produced several early ski pioneers such as Georg Bilgeri and Hannes Schneider, who were instrumental in advancing ski technique and developing pedagogical methods. Schneider’s methods in particular were extremely innovative– most people worldwide today still learn the Arlbergtechnik that he developed. Bilgeri and Schneider would use their innovations to revolutionize skiing in the Austro-Hungarian military, and with the outbreak of WWI, they were responsible for teaching soldiers how to ski in preparation for the “White War”– the war for the contested mountainous border between Austria and Italy. After the war, skiing entered the realm of the masses, and winter tourism in Austria– especially western Austria– skyrocketed. Tourism hot-spots such as St. Anton in the Arlberg region helped get Austria through its post-war depression, and Hannes Schneider continued to draw people to both St. Anton and the sport of skiing through his work in the Bergfilm industry. Today, Vorarlberg is still well-known for its world-class ski resorts, and many Vorarlbergers are put on skis soon after learning how to walk. Austrians went from fearing the mountains at the fin-de-siècle to hosting the winter Olympics in Innsbruck in 1976, attesting to the power of the mountains and the incredible innovations from Vorarlbergers that made mountains both safer and more enjoyable for recreation
Taking Sovereignty to Court: An Inquiry into the Trial of Bahadur Shah Zafar
Full text linkMuhammad Bahadur Shah Zafar – the last Mughal emperor – was tried for treason and sentenced to exile within a British court of law at the conclusion of the Sepoy Mutiny in India in 1857. The trial decisively brought the Mughal Empire to an end while fundamentally transforming the nature of colonial authority in India, from Company Rule to Crown Rule. Many scholars have argued that the trial was conducted as an “afterthought” to the violent repression of the Mutiny by the British forces – a formality in the inevitable establishment of the British Empire in India. This view offers a limited understanding of the trial and the role of the law in shaping sovereign authority in India in the nineteenth century. This paper argues that the trial of Bahadur Shah Zafar played a crucial role in restructuring the colonial episteme after it had been violently challenged by the Sepoy Mutiny. This colonial episteme, which emerges through the trial, allowed for the simultaneous construction of Mughal treason and justification of British claims of sovereign authority in the Indian subcontinent. The restructuring of the colonial episteme did not entail a dismantling of pre-existing imperial networks and structures in the Indian subcontinent. Rather, by enfolding these networks into the colonial episteme, the trial played an important role in transforming these networks to justify British claims of sovereign authority in India. To elucidate this argument, this paper aims to critically analyze three types of evidence that were presented during the trial, namely: documentary evidence, testimonies, and newspaper reports. Finally, this paper argues that, by invoking the discourse of the law, the trial of Bahadur Shah Zafar restructured the colonial episteme, which, in turn, shifted the locus of sovereign authority within the Indian subcontinent from the Mughal sovereign to the British Crown
Investigating the Pathways from World Trade Center Exposures to COVID-19 Severity and Long-Term Sequelae
No full textOur study examined the associations between WTC exposures and COVID outcomes using two analyses: one focused on COVID severity and the other on post-acute sequelae of SARS-CoV-2 infection (PASC), which is known as long COVID. In the severity analysis, our aggregate exposure measure revealed that a significant portion of the effect on COVID severity is mediated by obstructive airway disease (OAD), indicating that impaired respiratory health is a key pathway linking WTC exposures to more severe acute illness. In the PASC analysis, we found that the overall effect of WTC exposures on long COVID is largely mediated by impaired respiratory health. These findings suggest that while OAD is a critical mediator for both COVID severity and PASC, additional mechanisms likely contribute to long-term outcomes
A System-Agnostic Approach to Computational Ideology Detection: An analysis of Colombian congressional Speech (2000-2024)
Full text linkThis study proposes a system-agnostic, modular approach for detecting ideological structures in political discourse without relying on predefined ideological axes or external labels. Building on a sociological understanding of ideology as the rationalization and narrative construction of political action, the method combines topic modeling, narrative mining, and frequent itemset analysis to inductively reconstruct ideological patterns from large-scale political speech. Applying this pipeline to a corpus of Colombian congressional interventions (2000–2024) revealed three major attitudinal stances toward peace processes, alongside narrative bridges linking economic governance, institutional trust, and post-conflict reform. These findings demonstrate that ideological structures can be systematically detected through emergent speech patterns, offering a flexible and culturally sensitive alternative to traditional ideology detection models. The results highlight new possibilities for computational social science in fragmented political contexts, particularly in regions like Latin America, where labeled data is limited and ideological systems are fluid
"They Forgot About Us": The Experiences of Dual Language Teachers in Southern California
Full text linkThis study explores how administrative practices shape the experiences of dual language teachers at two elementary schools in California, Grove and Orchard. Drawing from 12 teacher interviews, the findings reveal school administration to be the most salient factor influencing teacher satisfaction, resource accessibility, and the faithful implementation of dual language programs. While both schools operate under the same district mandates, administrative behavior mediates the impact of those policies in distinct ways. Orchard’s administration demonstrated greater support for Spanish instruction and teacher collaboration while Grove’s administration exhibited a preference for English and offered limited institutional support. These differences affected teacher autonomy, resource allocation, and coworker collaboration. Additionally, the study reveals how district-level decisions and inadequacies in bilingual assessment tools perpetuate inequities and undermine the legitimacy of dual programs. The findings underscore the need for administrators and districts to engage meaningfully with dual language programs and teachers to create equitable and effective bilingual learning environments
Circles of God: Frame Drums and Ritual Soundscapes in the Middle East
No full textThis thesis examines the frame drum as a central instrument in ritual practices across the Middle East, focusing on how it operates within ritual niches—distinct spiritual and socio-cultural settings that give instruments their meaning. Through case studies including Sufi dhikr ceremonies, Kurdish rituals, Yazidi Qawwal performances, and the Fjiri music of Gulf pearl divers, the project explores the varied roles of the frame drum across devotional and communal life. Rather than engaging in detailed musical or technical analysis, the study emphasizes how the instrument mediates relationships between memory, theology, and embodiment. It argues that frame drum practices are shaped by regional histories of cultural contact and exchange, and that their divergent uses across communities reflect broader patterns of transculturation within ritual life. By highlighting understudied traditions and cross-regional connections, this thesis contributes to a fuller understanding of how frame drums help structure spiritual experience and ritual authority within the Islamicate world
Navigating Action Crisis: Identifying Effective Support Types to Sustain Goal Pursuit
Full text linkThis research explores how individuals in different stages of goal pursuit respond to various types of social support, with a particular focus on those experiencing action crisis, a motivational conflict about whether to persist or abandon a goal. Across two studies (N = 200), participants evaluated six types of support (evidentiary, value-focused, active listening, efficacy, instrumental, negative validation) in terms of perceived genuineness, self-verification, and emotional response. Study 1 presented hypothetical goal scenarios and scripted support messages, while Study 2 increased personal relevance by asking participants to reflect on a real goal and imagine support from a trusted close other. In both studies, support type significantly influenced participants' perceptions, with efficacy support consistently rated favorably across outcomes. Contrary to hypotheses, empathy-centered support types, especially negative validation, were not rated more positively during action crisis. Negative validation was rated lowest in Study 1 but received more moderate evaluations in Study 2, suggesting the tone and relational context of support delivery may influence its reception. Goal pursuit condition (implemental vs. action crisis) did not significantly moderate the effects of support type in most outcomes, though Study 2 revealed a significant interaction for self-verification, suggesting subtle interpretive differences based on motivational state. These findings highlight that while empathy may contribute to positive support experiences, confidence-boosting strategies like efficacy support may be equally or more effective, especially when delivered by someone the recipient trusts. The results underscore the importance of support tone, delivery context, and relationship closeness in shaping support effectiveness