Academy of Media Arts Cologne

Phrases like “This prompt violates our content policy” or “As an AI model, I cannot...” indicate the invisible edges of generative AI systems. These edges are both frustrating and tempting. The concern persists that the massive scale and statistical structures of large models leave them impenetrable and incomprehensible. This paper examines large language model guardrails as a case study for applying critical code studies methodologies to large-scale AI. The subfield of AI alignment provides a rich access point for analyzing how foundation models are moderated and refined. This study explores four popular companies’ guardrails—Anthropic, DeepSeek, Meta, and OpenAI—including both how guardrails are applied in general-purpose models and also how they are built into moderation API tools on public offer. The analysis looks at endpoint documentation, code examples, technical reports, model architectures, training dataset content, and methodology research to create a critical picture of guardrails as part of conversational interfaces for sociotechnical control. It maps how their technical construction also co-constructs ideology through language, both computational and linguistic. Through guardrails, certain conversations are defined and limited by their filters, while other conversations are promoted. Thus, code decodes, encodes, regulates, and is itself conversation about what can be discussed.

This study employs a reflexive thematic analysis to systematically examine the design patterns of 48 first-person Virtual reality (VR) animal avatar games. The research identifies four primary design themes: Animal Biomimicry, Limited Animal Simulation, Hybrid HumanAnimal Features, and Human Behavior with Animal Avatar. The analysis reveals that approximately 77 percent of the games remain grounded in human-centered interaction logic, with animal forms primarily serving as visual representations. The study highlights the core tension between authenticity and usability in current VR animal avatar design, and points toward design opportunities for achieving more authentic animal avatar's interactive experience through directions such as controller innovation, unconventional body mapping, and dynamic feedback. This research provides a thematic classification framework for understanding the representation of non-human perspectives in VR games.

This article investigates the subversive potential of skateboarding, examining how a practice long associated with urban masculinity can be reinterpreted through a queer theoretical lens. While mainstream skate culture – exemplified by outlets such as Thrasher Magazine – has historically reinforced heteronormative and male-dominated representations, the article reads the activity of skateboarding itself as destabilizing conventional notions of gender, embodiment and spatial engagement. Drawing on Jack Halberstam’s theories of queer failure and queer temporalities, alongside Hannah Freed-Thall’s concept of queer ecology, the analysis highlights the symbolic and material importance of empty swimming pools in skateboarding history. These repurposed spaces become sites of alternative movement and meaning. A close reading of Tara Jepsen’s novel Like a Dog further illuminates skateboarding’s queer dimensions: through its protagonist, Paloma, a dedicated pool-skater, the novel shows how practices of falling, failing and appropriating space foster moments of resistance and alternative ways of relating to the world. Ultimately, the article argues that skateboarding has always transcended its masculinized origins, offering distinctly queer possibilities for inhabiting urban environments.

Published in 1966, Joseph Weizenbaum’s ELIZA is the first conversational agent or chatbot, most famous for its DOCTOR persona. In 2021, a group of scholars recovered the original source code for ELIZA from the MIT archives. For about four years, an expanded team of eight scholars from diverse disciplines explored this code through weekly meetings in one of the longest ongoing projects of critical code studies. This paper presents both the methodology of conversational close reading as well as the substantive findings that resulted from the sustained collaborative inquiry. Their investigation examined not only how ELIZA functions technically but also what it reveals about the cultural moment of its development and its enduring influence. The team interrogated the code by reading it line-by-line, porting it to other languages, and attempting to run it on emulators, while also consulting archival materials and conducting interviews with Weizenbaum’s contemporaries. This collaborative approach revealed significant discrepancies between Weizenbaum’s published descriptions of ELIZA and its actual implementation, including previously undocumented features and editing of canonical conversations. The project demonstrates how critical code studies can operate as a conversational method, where a team of scholars with diverse disciplinary backgrounds can use a technical analysis to open up a broader cultural and philosophical discussion. By treating source code as a cultural text rich with historical significance, this work bridges software history with code analysis and offers insights into the development of human–computer conversation that have become more increasingly relevant in today’s era of large language models. This paper selectively documents the team’s process, discoveries, and reflections, offering their collaborative model as a template for understanding both historical and contemporary computational artifacts for future projects in critical code studies.

In diesem Beitrag gibt die deutsche Lyrikerin Monika Rinck einen Einblick in ihren jüngsten Gedichtband Höllenfahrt & Entenstaat, der 2024 bei kookbooks erschienen ist. Er erörtert und bezieht sich auf das Leitmotiv der Höllenfahrt und ihre verschiedenen Erscheinungsformen in der urbanen Gegenwart. Dabei entfaltet sie anhand einiger Gedichte der Anthologie, wie Poesie imaginäre Räume schafft und verkörpert.