3 ms·
Cool! How does it work?
by bjourne 4y ago
Cool! How does it work?
- rpastuszak 4y agoThe page and the repo contain more context (give it a read, even if just for the buttery puns!). But, in short: It's quite crude. I extract captions from YT videos using a private YT API (web scraping was a huge pain ;/), then I massage them a little bit and use GPT to process the timestamps. A waaaay better approach is possible, but I'll need to see if it's worth pursuing first. PS. SponsorBlock has a huge timestamp DB so I'm considering reaching out to the authors to see if we can create an agent using their extension and aiding their crowdsourcing effort.
- FloatArtifact 4y agoAwesome project! I'd love to see that