I've always been interested in AI, but never took the time to learn it. This time, I wanted it to be different. After creating a simple model in Tensorflow to recognize hand-written digits from the MNIST dataset, I started learning how to create an RNN.
The idea was to take my personal Instagram DMs and train a model on them to generate text. Seeing as how I'm a complete noob when it comes to machine learning stuff, most of the model I wrote is just my interpretation of Tensorflow's text generation with an RNN tutorial. I barely understood any of the terminology, but Google came in handy.
Here are some of the results of the trained model! The results have been heavily curated since even though they're unique phrases, they can still sometimes contain private things.
doesn't even take it request of Spone.?!?!
Like what website?
Lol
Oh yeah
Spent you do this time
From the spetify like, that it would be faster
AHHH
oh
Someone shopp day tbh
As you can see, smaller more common phrases are replicated perfectly. However larger sentences make no sense. They can be funny though. You can even see some basic grammar rules like contractions and commas. This little experiment was really fun to do, if not infuriating at times.
Top comments (2)
Great article Ryan, I may have to try that :)
Thanks Fulton! I might open source it later.