This Word Does Not Exist

AI generated English words with dictionary definitions

With every click, a new AI-invented word, complete with pronunciation and dictionary definition.
Discussion
Would you recommend this product?
3 Reviews5.0/5
Both fun and intriguing. Would love to learn more about what's happening under the hood.
@davidbauer Thanks for hunting! What do you want to know more about?
This is a fun little product for sure!
Maker here! Thanks for hunting; it was a ton of fun to build and evolved out of a conversation about potential AI company names.
@thomas_dimson It certainly made me want to purchase a few domains :)
@thomas_dimson (you could probably monetize it that way too)
@davidoudiette I suppose so! Or baby names, company names and that kind of thing. Right now this is a hobby project for me
Hey, this is clever! Impressive you were able to automate something language-oriented like this. Any favorite words it's generated so far?
@maxmirho1 honestly, it's been my lifelong dream to define some of the words made up in the Simpsons. It's neat to see a real definition for kwyjibo https://l.thisworddoesnotexist.c... Otherwise, I can always recommend trying to define yourself. Apparently I'm a classic North American fruit tree https://l.thisworddoesnotexist.c...
@thomas_dimson Hell yes dude :) I love these, I can't wait to plug a ton in.
Hey, I am bit curious. How does the AI even generates the word's meaning as well?!
@realdesigntack internally I represent definitions as a sequence of tokens |START| word |POS| noun |DEFINITION| a word definition |EXAMPLE| the example usage |END| From there I train a model that learns how assign a probability to those types of sequences. Essentially, it tries to predict the given what it is seen so far. You could say "d" is likely to follow "|START| wor". The meaning and example usage is captured in this probability -- a more plausible meaning is more likely than a crazy one (hopefully). After I have the probabilities I generate one token at a time by throwing the (weighted) dice one character at a time.
@thomas_dimson Woah, that sounds complex yet interesting. Good job, mate.