Disrupting the hobo naming industry.

View the Project on GitHub raritet/i-hobo


Noted humorist John Hodgman had this thing where he named seven-hundred made-up hoboes in his 2005 book The Areas of My Expertise. He read them all aloud for the audiobook, which apparently took an hour. You can read up on the whole story, but it became an early internet meme with various remixes and offshoot projects including the now defunct e-hobo.com where one could find illustrations of the various hoboes. (Note that at this time a “meme” wasn’t specifically an image with some text on it.)

Somewhat late, maybe ten years late, I added to the meme, drawing on some new silliness of that later time. (Can I call that the Albernheitgeist?)

So, inspired by the AI Weirdness blog, I fed Hodgman’s nice, ready-made corpus of 700 hobo names into an LSTM neural network in an attempt to train it to generate its own hobo names, thereby disrupting the hobo naming industry with all the buzzwords. It was basically a script kiddie operation, repurposing some Lua code and futzing around to get library dependencies sorted out.

I set up a Twitter account (that I have since abandoned) to tweet these names regularly along with registering the domain i-hobo.biz (that I have also abandoned) to host a little write up on the project.

It worked OK. We didn’t have GPT-3 and stuff like that back then, or at least I didn’t, so the brain was pretty simple, with each training session lasting only a few minutes on a mid-range 2016 MacBook. But even then, the results sound like they make sense even though they almost always don’t.

Either way, I find these things endlessly entertaining. Here is a raw dump of the last batch I made. I highlighted my favorites.