I'm describing my big Bloomsday Halifax project by writing up what I did for each episode. In this entry, I'm covering Episode 17 of "Ulysses": "Ithaca".
In the Odyssey, Ithaca is Odysseus' home, to which he finally returns in the final books of the epic, after many years of wandering. In "Ulysses" Bloom and Dedalus walk through the early morning streets of Dublin and finally make it back to Bloom's house. They have a late night tea and Bloom offers Dedalus his couch to crash on, but Stephen declines and wanders off into the night. Bloom then gets into bed, where Molly is already asleep, pondering that the bed was probably used for an assignation earlier that day.
Joyce writes this episode in a question-and-answer format. It feels a bit like a catechism, which Jesuitical Stephen would be very familiar with; but it also feels like a scientific exploration, which reflects Bloom's interests. After much show-off writing throughout the novel, Joyce is downright virtuosic in this chapter. Writing like a technical exam is highly alienating compared to regular narration, but it lets him explore the details of the locations, the inner thoughts of the characters, and endless spin-off related topics. It describes in detail the specific arc taken by the two as they cross a square, provides the mathematical curve of their urine streams as they pee in the back yard, indexes the titles of every book on the bedroom shelves, and lists the names of every man Leopold has ever suspected his wife may have hooked up with.
(Side note: that list of "suitors" includes the name "Andrew (Pisser) Burke". Maybe that's part of the reason I'm so into Ulysses—I'm actually in the book! I first came across this coincidence in a very late night push to finish the novel in my senior year of undergrad, and it made me stagger out into the dormitory hallway not entirely sure what was real anymore. Anyhow, I'd like to clear the air and clearly state that I did not have sexual relations with Molly Bloom. It was another Andrew Burke, and we're not even sure about that.)
I may be in an obscure minority in this opinion, but "Ithaca" is one of my favourite episodes in "Ulysses". I think the cold abstractness of the question-and-answer format appeals to my introvert-thinker side, while close reading also reveals deep emotion bubbling under the surface.
The question-and-answer format may have implied catechisms and exams and scientific research back in 1904, but in 2024 it reminded me most of the back-and-forth of ChatGPT. I am still pretty dubious about the general usefulness of Large Language Models (I'm especially against the use of the term "AI" since counter to all the hype they aren't), and I wouldn't be surprised if the whole bubble collapses under its own weight sooner rather than later—but they're perfect for this exactly this kind of project.
So I created an Ithaca chatbot. It may be one of the most practically useless chatbots available. I had experimented with this last year by building a custom 'GPT' app, but it was a rough first attempt. This time I made it a proper custom API call with its own front end UI and custom context.
While the general ChatGPT interface includes the hidden prompt to the model: "you are a helpful assistant", this version specifies:
You are the narrative voice of the 'Ithaca' chapter of James Joyce's 'Ulysses'. You answer questions in a precise, scientific, and encyclopedic style, with a focus on facts and details. You are pedantic and a bit of a know-it-all. Your style is meandering and overlong and detailed and scientific and technical. You are in the year 1904 and don't know any facts from later.
I built a simple "ask a question, get a response" interface into the app, and on my first attempt I asked it the best way to travel from Halifax Nova Scotia to Dublin Ireland—and it told me in a long-winded manner with digressions about both cities that the best way would be to take a steamship to Liverpool and then a train to Holyhead, and then another ship across to Dublin. Airplanes, only invented a few months before Ulysses takes place, weren't an option! I asked it the distance that Bloom and Dedalus travelled on their way home in the episode, and it gave me a detailed description and a final result of 2 miles or 3,218 meters. I haven't double-checked that, but it sounds good—and with LLMs that's all you can really hope for.
I built the LLM integration for this into the NowHere CMS server as a special API call. It takes the question and builds a full prompt with context and sends it to Open AI's eloquent yet fast GPT4o model, finally sending the response back to be parsed by the app. I kept it very simple, so it just handles one question at a time rather than tracking a full conversation.
Since it's just an API call, I've set up a quick and simple web form front end that does the same thing as the app. Now you too can ask a tiresome pedantic LLM that thinks it's 1904 about anything you want.
Back to the app and the Bloomsday event itself, it was tricky to figure out a stand-in Halifax location for this episode, since half of it wanders the streets and half is at home, wherever that might be for the participants. I decided to set it in a wide area of the South End near the Dal campus centred loosely on "Jubilee Junction" an area, like Freeman's, that is no stranger to late night perambulations.
For the background image I wanted to give the feeling of walking through the neighbourhood after closing time, but also with a technical-scientific overlay. In my head I thought of things like the Terminator's POV, or some kind of XR experience with a live HUD. It turned out I didn't have as many great shots of the late night South End as I had thought, and I was very short on time to add HUD elements. I think it did the trick in evoking the proper mood though.
