If LLMs had any degree of accuracy guarantee then they would be useful.
I often think, “I could let an LLM do this and save me time, if it worked…”. But it’s always been a professional bullshit generator.
Can’t trust it to make even a simple list, so you always end up doing double work.
A PSA for you is that the original setup for local invidious is not really worth it at this time. You gotta rotate IP to avoid youtube blocking it, you need to use IPv6, and you have to constantly refresh PO tokens. All of this happens with separate programs so it’s clunky and can’t keep up with youtube’s rate of blocking things, unless you setup some kind of frequent cron jobs, but at the cost of bringing the service down frequently which is not worth it.
For my instance I use the companion setup. It takes care of rotating PO token by detecting when it needs to. It’s also way more robust, based on very well maintained youtube.js project.
Another change I made is routing the invidious connections through VPN. That keeps my real IP safe from getting blocked, and I do remember this setup rotates the VPN server when blocked. I’ll share my compose file in a bit, been using it with almost no issue for almost a year.
EDIT: https://pastebin.com/0RqR77i2