A live experiment in economic geography

An economist will write an academic paper in six days on a beach in Hong Kong's Lantau South Country Park, using only generative AI for assistance and livestreaming the entire process, to test whether prompt-based research can survive desk rejection at a Scopus-indexed journal. The experiment, starting with a ferry to Chi Ma Wan at 22°13′21″N, 113°59′03″E, will record all interactions with models to observe tacit knowledge relocation, and the results will be published regardless of outcome.

Writing an entire paper in six days from an island in the South China Sea, with no one but machines to ask for help and no one else to blame . Email me with updates on the project mailto:hello@disembedded.com?subject=Updates%20%E2%80%94%20Disembedded&body=Send%20me%20updates%20on%20the%20project. Two emails. One when the paper is finished, one when it is published or rejected. Nothing else. Can I write an academic paper that survives desk rejection at a Scopus-indexed journal? Lamb chops, marinated twice over 48 hours, in Central. Eaten alone, on camera, with no one to split the bill with. Reviewing, letters, discussant duty — whatever is asked, until the hours are paid. Also livestreamed, in full. The bigger things this project is about, besides the best lamb chops in Hong Kong. A longstanding argument in economic geography is that economically valuable knowledge is often tacit: difficult to codify and acquired through experience and practice. Generative AI seems to undermine this. Interacting with a model requires the user to explicate mental processes that would otherwise remain instinctive, and the model supplies codified answers at a distance. My hypothesis is that prompt-based research does not eliminate the importance of tacit knowledge so much as relocate it, for instance to recognizing that a question has been posed badly, deciding what to ask next, or detecting that a fluent answer is somehow wrong. By recording my interactions with models, I can observe behaviors that I am not thinking about at the time, and that I could not fully articulate if I tried. The footage also adds to one of the smallest archives in the world: tacit knowledge videos https://www.lesswrong.com/posts/SXJGSPeQWbACveJhs/the-best-tacit-knowledge-videos-on-every-subject . Coming on the heels of a replication crisis, AI has introduced an entirely new way for academic integrity — however defined — to be undermined. The movement toward open access and preregistration has productively made academic production more open. But once tools can stand in for so many parts of the work, full transparency may be the only stable frontier of openness left. The riskiest experiment here is the one about content. The creator economy has revealed pent-up demand for things once considered genuinely weird: people playing video games, people unboxing gadgets. Will academics and other weirdos turn out for live scholarship? This is one attempt at finding out. Lantau is Hong Kong's largest island, and it holds one of the most connected places on earth: the international airport, the bridge to Macau and Zhuhai, the rail line into Kowloon. It also holds Chi Ma Wan. The desk sits at 22°13′21″N, 113°59′03″E — a beach on the western shore of the peninsula, inside Lantau South Country Park , country park behind and open sea in front, on a shoreline that once stood in for somewhere else in a Jean-Claude Van Damme film. I take the ferry out to Chi Ma Wan and start recording. From that point I have no contact with my co-authors, my colleagues or anyone else in the research community until the recording stops. I open the broadcast by explaining what the paper is supposed to be and what I think is most likely to go wrong. Then I put the first question to a model on camera. I write during the day, from eight in the morning until six in the evening, Hong Kong time. The stream is unedited, so you will see the prompts I write, the answers I throw away, the dead ends I follow, and the long stretches where very little happens. These are casual hangouts with colleagues and viewers about academic writing with AI, and anyone watching can ask a question while we are live. Nobody helps me with my own paper. Slots still open — come on stream guest . I leave with whatever I have managed to write. It might be a finished draft, it might be half of one, and it might be a cautionary tale. I will publish it either way. All three of us watch the footage on our own and then compare what we found. Two of my co-authors take no part in the writing week, so their reading acts as a check on my account of my own process. Stay tuned to find out. Each evening I open up the stream for conversation with other researchers about how they use AI tools in their work, and to pick their brain about what they think the future holds. One rule: nothing substantive about my own paper. No feedback on the analysis, no suggestions about the data, no reading of drafts.