i greatly dislike how much the word "AI" is thrown around to apply to any sort of model generation but i feel like using offline language and speech-to-text models for automating a lot of the closed caption generation process for e.g. youtubers is pretty helpful, _if_ that tooling gives you some output that strongly suggests you watch the result and gives you really easy adjustments, and the default output is way better than youtube's automatic script-to-captions shit

but i also feel like you probably don't need to scrape the entire internet to build a model of how to naturally split sentences for captions. i feel like we had research on that before openai fired a spider gun at the internet

i imagine the reason youtubers don't actually check their autogenerated captions is because when they (or their editor) is done editing they don't want to look at the video again because they'll keep seeing things to tweak or fix

augh no now i’m mocking up a tool in my head for watching and fixing up captions I DON’T NEED ANOTHER PROJECT

@iliana I am also mocking up a tool in my head, somehow your brainworm has infected me as well! dangerous