Hello. I used to be a scholar in AI, anti-racism, and religion. Now I'm just a guy with a personal website.

Closeup of Will Penman
Current Projects

My hobby projects currently involve web design and ChatGPT prompting:

  • haikus.news - haikus written by ChatGPT every morning based on the day's top stories, with links to the corresponding New York Times articles. I'm delighted by this project. As a user, I enjoy waking up to new haikus. I love having poetry as an intermediary between our normal mode of being, versus the stiff impersonal language of news discourse. At a technical level, I sometimes tweak the prompts to push ChatGPT to its creative limit.

  • End-to-end novella generation - I've been using my writing teacher skills to prompt an LLM to write an entire novella end-to-end, using just a 2 paragraph input about the plot and themes. This project began in November 2023, when for NaNoGenMo I wanted to write an "in principle automate-able" novella using a story idea I had in college that I never pursued - an AI that gains consciousness and then sees it fade away. I crafted a chain of prompts by hand that each helped design the story. ChatGPT's output from one conversation would determine part of the input for the next conversation, and so on. I ran these prompts through the ChatGPT website one by one in 60 conversations, but that only covered the planning phase. To finish that story, my brother Andrew and I automated the drafting step using the OpenAI API. At the end of May 2024, we successfully generated The Resonance Experiment, a proof of concept for a novella writing system. Encouraged by the results and seeing a lot of room for improvement, we decided to continue with the project. In early September 2024, we completed Frostpearl, our first full prototype, demonstrating an entirely automated "one-click" process that also displays much stronger writing. Frostpearl is about 30,000 words, putting it right in the mid-range of novellas (stories you can read in one sitting). At a technical level, our Prototype 1 system is entirely 0-shot and mostly single-sample, with no agentic components. To achieve a high level of narration, we've been innovating in expert-driven task decomposition. Specifically, we a) create a simplicity-first modular workflow (using only 8 writing tasks in this prototype); b) hand craft prompts with scaffolded task achievement; and c) diagnose areas for improvement using writing expertise rather than automated evals or automated prompt chaining.

  • How To Prompt ChatGPT Like A Science Writing Pro - A downloadable guide of prompts for students to integrate ChatGPT into their article drafting workflow. (A professional prompting project)

  • Dog Birthday Calculator, created during Covid - My first website to use extensive front-end calculations. It's a fun one-off site that fills a need, since no other site projects out to the future so you can celebrate your dog's "dog birthday." Currently, the site is over-engineered; I intend to scale it back and remove a lot of the customizations.


About

From 2018-2022, I was a Lecturer at Princeton in the Princeton Writing Program. I taught first-year undergraduate students how to design and conduct small-scale experiments about AI and then write about those experiments scientifically. At the beginning of the semester, students' experiments were focused on identifying social bias in GPT-2's output (state-of-the-art open-source LLM at the time). By the end of the semester, students selected their own AI-related research to investigate and write about. These varied by scholarly field and sophistication, but were sometimes extremely successful. For instance, one student, Arti Schmidt, compared algorithms for autonomous drone racing, and then he published it based on his work in my class. Another student, Byulorm Park, studied artists' resistance to AI face detection, and developed a theoretical framework in an independent study with me. Her essay, "Countervisuality for a Gentrifying City Center: Structuring Systems of Surveillance through Architecture," was awarded runner-up in a national research contest meant primarily for Masters students.

Now, I run a small business to coach PhD students on their writing: Composition Coaching. Scientific writing is hard and PIs often aren't trained in writing themselves, so I offer guidance, encouragement, and accountability to help students finish their articles and big projects with less stress.

I also pursue some projects as an independent researcher. My 10+ years of interdisciplinary work from teaching/coaching allows me to be nimble, insightful, very organized, and unafraid to tackle detailed projects.

Before I taught at Princeton, I earned my PhD in Rhetoric at Carnegie Mellon University. I currently live in Pittsburgh with my wife and our son.