Good question. You can contact us on Twitter @yourmumdotsex or email us at [email protected] if you want information about the dev team.
This site uses OpenAI's GPT-3 to generate jokes about your mother. We've fed the system with as many jokes as we could come up with in order to achieve maximum destruction in your next argument.
GPT-3 is what's known as a neural network-powered language model. A language model is basically just a program that can calculate the likelihood that a random set of words is a valid statement. For example, given the choice between "I took my dog to the park" and "I took my house to the park", it would be able to determine that the first option is a more correct statement.
GPT-3 was trained on a massive library of data, including the entirety of the English Language Wikipedia and CommonCrawl. Words and sentences were randomly removed from the content of passages, and the language model needed to fill them in correctly based on the surrounding content.
This is basically a comprehension exam for primary school students, but instead of 40 questions, it's effectively the entirety of human knowledge being tested on. As a result, the final model is able to accurately generate text in all sorts of fields, not just one specific area.
GPT-3 is a transformer-based neural network, which means it looks at each word or phrase and determines its significance in a sentence. For example, in the sentence "I took my dog to the park", the words "dog" and "park" are probably the most important, followed by "I" and "took", with the rest of the sentence being essentially context clues. Transformers are able to take in an entire sentence or paragraph at once, rather than processing one word at a time. GPT actually stands for Generative Pre-trained Transformer!
A lot of really cool stuff is possible as a result of the generalised nature of GPT-3. I've used it in the past as a grammar corrector, an English to French translator, a text summariser, and have even written entire assignments for school using it (Fun fact: GPT-3 can pass HSC English Advanced!). Other people have used it for more useful things, such as swearing filters, imitating famous poets, and coding entire projects using under 10 lines of instructions, something which would have been impossible a few years ago.
This blog post from Towards Data Science explains the way GPT-3 works quite well too.
Basically, we've fed GPT-3 a whole load of your mother jokes, and then asked it to come up with new ones. In theory, every time you enter a starter into the text box above, it should generate a brand-new joke about someone's mother.
Not yet, we're doing a lot of weird workarounds for things in the backend, and our code documentation is kind of non-existent. If you're interested anyway though, drop an email to [email protected] (no I'm not kidding that's the email) or Tweet at us @yourmumdotsex and we'll get back to you.
We might open-source it eventually, but we're not quite there yet.
We have a Liberapay page, because they don't take any fees (unlike Patreon). You can donate as much as you'd like, everything helps with the hosting costs :)
Alternatively, you can use the submission form here to give us more training data. All you need to do is full out the form with your funniest jokes,
You can follow us on Twitter @yourmumdotsex to get progress updates,
Or you can just share this site with your friends!