Work with Dearrow to improve reducing clickbait

Thibaultmol

Regarding the recent the recent
"Protecting young minds: Enhanced video safety for families"
from the changelog (oct 24 2024)

I think that it would be really good if Kagi had an integration with Dearrow so it can leverage it's massive database of curated titles and thumbnails for videos to reduce clickbaits.

Because it's an open source project, it would be nice if kagi could financially compensate it's usage of Dearrow to some degree then tbf

https://dearrow.ajay.app/

This should be a setting btw, let users choose what kind of changes they want to happen, similar to the dearrow extension settings

ni_something

I read Kagi's latest release notes and was thinking the exact same thing :-)

Thibaultmol

Good point brought up by mojolobo on the Discord: That content is licensed CC BY-NC-SA so can't be used by Kagi.
Unless Kagi manages to get a special license deal or something with Dearrow (if they're even able to do that)
https://discord.com/channels/1256077108111868035/1262752714757247037/1299443665130750012

carl

I think this is all backwards. Shouldn't Kagi surface high quality video results, instead of trying to mask low quality results by changing the thumbnail? If a video creator has a clickbait thumbnail, I know to not click that video. So the thumbnails serve a purpose.

Thibaultmol

carl I mean, I agree. But how are you going to determine what a good quality video is

carl

Some kind of analysis of the transcript is probably the way, maybe coupled with AI analysis of some snapshots from each video. But to do this, YouTube would need to have such an index, or you'd have to make one. Or maybe bypass YouTube's search API completely and source YouTube videos from links in a general web search.

If somebody opens YouTube for entertainment purposes, to find something to watch, then I understand why you'd want to sanitize thumbnails. But if I'm searching for information, I'd like the original thumbnails to remain, so that I can know to avoid videos with clickbait thumbnails, since the video itself will be low quality.

Thibaultmol

carl I can't see it being feasible to ai analyze so many videos though