I’ve trained this open-source neural network framework on a variety of datasets, including recipes, Pokemon, knock-knock jokes, pick up lines, and D&D spells.

This time, the dataset was a list of almost 2000 car models, helpfully provided by readers Anya and Dan. Can the neural network learn to suggest plausible car models? (Spoiler: Well, kinda…)

The neural network had to start from scratch, figuring out first what line breaks were, and then how often it should use them (early iterations consisted of long streams of line breaks and nothing else, the network apparently having figured that since it couldn’t yet match the rest of the data, at least it knew how to do line breaks).

Then there was a strange stage where every car name sounded like it was invented by cats:

Maru a
Marr
Mauu a
Mauu e
Mauu
Maru e
Mauu

Followed by another stage where it had figured out how to use other letters and numbers, but not how to use them wisely. It ends up sounding less like cars and more like early internet screennames:

Fatiuln Fo NT-Fi0i
Defruti Xo5m
Trfamlut Coe0r
Mercuren-Be B-Toqu
Fozurn G0tcakgn Mabuttsoee E
FLlp0oo2dr ST4rr

We proceed from there to what I like to call the Dracula phase, where it begins to try to spell some of the major car makes, but ends up (mostly) sounding sinister instead:

Morcula Sapira
Mercult Sarii
Deriula Framno
Moroa 3-S
Darcult Sar
Mlrru Pie
Nunsgan Caakes

It continues to learn, and now we have knock-offs of the major brands, the sort of fine car you might buy at a very, very discount car lot:

Mosseni-Binz C SL5
Perrari 9r0
Sugkswagen Cepgyst
Posda Suzek
Subaau Veciro
Fonda Caxii
Nirkswagen Tiatukrr
Vinkswigen Xeama
Coyota Slnnae
ToyoMa 64

Finally, the neural network has grown more sophisticated, and has learned to spell the major car brands. It can generate the simple model names, the ones that are just a letter or two and a number. However, I would not put it in charge of marketing just yet…

Volkswagen Colon
Buick Shoat
Buick Crapara
Buick Apron
Fiat Deter
Fiat Coma
Fiat S-0-S
Fiat Doug
Subscribe now