NaNoGenMo draft conspiracy video generator

I recently published a search engine that indexes the subtitles of radical YouTube communities. For instance, if you search for a specific query, like "voter fraud", it shows up all the relevant clips.

For my NaNoGenMo submissions I've been thinking about creating automated YouTube videos around conspiracies, and see how much sense they still make. It's very much a WIP, I used Max Woolf his GPT-2-simple for the text generation.

I think one of the most important parts that I need to fix is to create clear starting points for the text.

Here's the script that I used to pull in the subtitles:

https://gist.github.com/evanzummeren/8e981a89d44ecd229cba63d0447ff43e

I then got about 1700 different subtitle files, that I merged in one text file through this command:

cat $(ls -t) > alltext

https://gist.github.com/evanzummeren/79aa0ada4023c82558f405832e8c2268