Startup Names

In 2016 I scraped about 67,000 names of funded startups off Crunchbase and trained a character-level LSTM to generate new ones. Two layers, 128 hidden units, roughly 330,000 parameters. You can train it on a laptop in half an hour.

You could do the same thing with a Markov chain, but there’s something about the way LSTMs learn language. The names come out hilarious and occasionally great:

Allo Pharms · Endara · Novical · Salution · VU Medion · Crork Grame

“Novical” sounds like it should be a SaaS company. “Crork Grame” does not. At higher temperatures the model gets drunk and starts producing things like “Flals Infynraftcoro,” which is not making it past any branding agency.

You can imagine the company that would have each of these names. The model doesn’t know what words mean, but it knows what startup names feel like — the truncated words, the vaguely Latin suffixes, the “-ify” and “-io” endings. It’s learned the genre so well that the output comes with phantom context attached. “Quantestria” is obviously a horse betting odds prediction service.

This was for Panda, an installation we did with OMA for the Oslo Architecture Triennale. The fiction was an autonomous AI that could create organizations as a form of resistance. It needed names for its startups, and the LSTM obliged. Tens of thousands of them.

The file sat there for years.

A digression. In 2021 I asked a transformer to write marketing copy for a content management system. It did, sort of, and then it hit the end-of-text token. The generator wasn’t set to stop, so it just kept going — no longer conditioned on the prompt, free-associating:

. You are limited to up to 250 comments per subreddits. Always use custom title if you want a different default image for a post. Headless CMS will use the JSON API version of the description. This support can be modified by changing headers

Will transform a todo into a list view with the ones important enough to press the Add button

Here are screenshots of development on my local PC (system running my Docker container)<|endoftext|>

Hey, look. A food truck, and an eatery that serves pies shaped like guns. Yay. This cross-over from cultiness to mainstream in just a few years: today, Jasper Street Bakery announces its second branch, in Elmwood Village. The new spot will span across two floors from the former Pizza Bakehouse, joining The Urban Poor, 394 Elmwood Ave., the recently-opened Falcata’s-themed Ray’s Italian Seafood Sandwich House, the impending Vortex (whose name pretty much says it all, tbh), and this very blog post. Chef Ravi Mathur (formerly of John Brown Smokehouse) and business partner Anil Gupta — who had come up with the name for the popular city institution back in 2010 — will head the place, according to the release.

“A Meat Cannon will compete for the hearts and minds of those who walk by,” tweeted prominent food critic Craig Claiborne, a.k.a. one of the last souls-with-food to call Jasper Street Bakery “the best place to eat in town.”

It goes on. Alfredo Laue, founder of Rotary Young Professionals, tweets about smoked meat pizza in Montreal. Then the model hits another end-of-text token and pivots into a Polish translator’s memoir about forgetting English gestures.

None of this is real. Craig Claiborne died in 2000. The model is hallucinating with genre — it knows what a restaurant opening press release sounds like, what a food blog sounds like, what a quote from a “diehard fan” sounds like. You could not get this out of a current model.

So anyway. I’ve had this pile of names sitting in a file and then the other day we actually needed one. Simen has been building an agent coordination framework — kind of like Slack for agents — called Miriad, and it needed a name.

I reached for the file.

You can stuff 30,000 LSTM-generated names into the context window of a modern model and have it find the ones that match a set of themes. There’s a lot of noise in them there hills, but the LSTM makes shit up and the LLM decides which shit is good. The misspellings are the best part — unique, with available domains, and occasionally genuinely great. Spelling ceramics with an X. That kind of thing.

Miriad was one of the names from the original set.

I’ve had Miriad redo the original experiment from scratch. The code is here: evenwestvang/lstm-startup-names.