Replies: 2 comments
-
I have a script that breaks down the YouTube or input text into chunks and runs the fabric pattern through each chunk in sequence as a separate item. This works if you want to treat each chunk separately but it does not understand the overall context. I am sure running it through the same chat session would achieve this but I haven't taken it that far. |
Beta Was this translation helpful? Give feedback.
-
The fabric -y switch has a secondary swith, --transcript-with-timestamps, that can be used to sort your problem out. The output will have the transcript for the video segmented by small 5-ish second windows. You can use perl, python, or awk to parse the transcript and select the sections you want (probably according to the video chapters you are after). You could try using the transcript with the timestamps, sent to an LLM (if you don't want to mess with lang's to statically get the data), but you'll likely need a powerful LLM, meaning you probably want to use your Vendor models with API keys (like Claude, Open AI, etc). If you want to accomplish this locally, and have problems creating the commands to do it, let me know a specific video and section you are trying to get, and I can try and demonstrate the process. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Hi everyone
So i consume long form youtube videos but sometimes its hard to sit through and id like to read through them
While i know there are models with large context windows most of them have a limited output limit
Is there anyway to use the "yt" function to grab only specific parts of the youtube video maybe following the chapters of the video and process one chapter at a time to get the most information out of these videos, i dont mind running multiple commands.
I have a custom prompt i use to get a breakdown of videos , while it works well for short videos i get very basic level output when i use it for long form content.
Beta Was this translation helpful? Give feedback.
All reactions