Mentored Research — Katherine Elkins

190+

projects

125,000+

downloads

4,700+

institutions

198

countries

Selected firsts

Across more than 190 projects in three Kenyon courses (Programming Humanity, AI for Humanity, and the Senior Seminar), one pattern repeats: a new capability ships, and the students Elkins and Jon Chun mentor are auditing it, fine-tuning it, or building with it the same semester. In many cases documented below, broader research communities formalized similar questions one to three years later.

See how the educational model works →

Frontier-lab humanities, before it had a name

In June 2026, The New Yorker’s “Eight Predictions for the Future of Higher Education” named frontier-lab, project-based research as one of the forms higher education will take by 2035. Elkins and Chun have run that model since 2016.

Timeline of firsts

Research chronology

Selected projects, anchored to what existed (and what didn’t) when students did the work. Every name links to the full paper on Digital Kenyon.

2018
Before GPT-2 existed Neural networks entered the humanities classroom through sheet-music generation (Seth Colbert-Pollack), style transfer (Miles Shebar), and reinforcement learning for trading (Tucker Bennett et al.).
2019
GPT-2 fine-tuning, the year it shipped Students used the 345M model for creative authorship, including Sex and the City writers’ room and centaur screenwriting, while others worked with StyleGANs.
2020
Gwern-cited GPT-2 work and GPT-3 Three GPT-2 poetry and lyric papers were cited by Gwern, including Jonah Zitelli, Kaiya Case, and Sophie Barrio.
2021
Systematic fine-tuning experiments Rebecca Lawson ran hyperparameter studies on GPT-2 screenplay generation; Grant Holt tested AI history-writing; Sarah Groustra asked whether camp is computable.
2022
ChatGPT red-teaming within weeks Adam Blum completed one of the earliest systematic jailbreak studies before the semester ended; related projects examined GPT-3 and prompt engineering.
2023
Auditing GPT-4 the spring it shipped Students tested TurnItIn detection (Abigail Foster), LLM self-assessment (Annalia Fiore), and executive-call simulation (Dillon Cleary).
2024
Real human-AI interaction at scale Hannah Sussman topic-modeled 3,275 ChatGPT conversations; Richard Anthony Álvarez built a RAG film recommender; Nava Bahrampour studied disability language and AI.
2025
Multi-agent systems and domain fine-tunes Godwin Idowu studied deception in multi-agent negotiation; Juliette Lowe fine-tuned BioLinkBERT; Parker Gibbons used agent debate for baseball decisions.
2026
AI agent networks Muhammad Ibraheem Nadeem analyzed Moltbook as an AI agent social network; Ayesha Aslam studied AI influencers and attention.

Browse the research

More than 190 projects, 2018–2026

Across three courses. Every title links to the full paper on Digital Kenyon; the selection below is representative. Projects flagged first were conducted in the same year the underlying capability shipped, or before the research area had a name.

Generative AI & Creative Writing Model Auditing, Red-Teaming & Evaluation Sentiment Analysis & Shapes of Stories Computational Text Analysis & Topic Modeling Social Media & Political Discourse AI Art, Music & Multimodal Creativity Human-AI Interaction & AI for Learning Agentic & Multi-Agent AI Applied AI: Finance, Business & Prediction AI Ethics, Law & Policy Health, Science & Society Capstone Builds

Generative AI & Creative Writing

Fine-tuning and probing language models for poetry, screenwriting, lyrics, comedy, and journalism, beginning the year GPT-2 shipped.

Model Auditing, Red-Teaming & Evaluation

Probing what models can and can't do (safety, reasoning, bias, detection), often before the research community had named the problem.

Sentiment Analysis & the Shapes of Stories

The program's signature method: computational analysis of narrative and emotional arc, applied to literature, translation, memoir, and film.

Computational Text Analysis & Topic Modeling

From Supreme Court opinions to the Septuagint: structure-finding in large textual corpora.

Social Media & Political Discourse

Computational social science on live political events: elections, protests, and public opinion as they unfolded.

AI Art, Music & Multimodal Creativity

GANs, style transfer, generative music, and computer vision for visual culture, beginning in 2018.

Human-AI Interaction & AI for Learning

How people actually use AI, studied empirically, and AI designed around human needs.

Agentic & Multi-Agent AI

Emotion, deception, debate, and network structure in systems of interacting AI agents.

Applied AI: Finance, Business & Prediction

Machine learning and LLMs for markets, forecasting, and organizational decision-making.

AI Ethics, Law & Policy

Algorithmic accountability before it was a beat: recommender systems, surveillance capitalism, IP, and emerging tech governance.

Health, Science & Society

AI applied to public health, medicine, and social crisis.

Capstone Builds: Products & Platforms

Senior Seminar projects that shipped: working platforms, tools, and systems.

Later uptake

Later use

Selected projects have been cited by Gwern Branwen and Meta AI researchers, featured by Forbes, and incorporated into later faculty research. For citation and adoption detail, see Scholarly Reception.

The project record remains available in the open-access Human-Centered AI collections at Digital Kenyon.