I built this to recreate those late-night Wikipedia rabbitholes. It uses AI to find two fascinating interesting topics related to a topic you choose and then lets you dive deeper.
Ok, but what do you need AI for? The two random topics don't seem very connected, they could just be random links (or backlinks) from the page. The potential is definitely there, but the execution could be improved.
They're certainly just random links from the page. I tested it on obscure pages with only a few links and it's very obvious that that's what it does. Take this page for example: https://wikidive.tulv.in/dive?topic=Alex_Alley
You can also go to this page that happens to have links to every date page and you can clearly see that it's not picking anything related to the article contents as most of the time it just picks a random date rather than one of the related topics: https://wikidive.tulv.in/dive?topic=List_of_non-standard_dat...
I don't know why the creator doesn't just say that it's random instead of making an easily falsifiable claim that it's "AI-guided".
OK, developer here. Here is how it works. It connects to groq and asks this
"I was going in this rabbit hole of wikipedia articles: #{chain} \n I have now found these topics:\n\n#{related_topics}\n\n Select the 2 most mind blowing topics from the list and that I'd enjoy given my rabbit hole. Ensure two topics are not the same."
and pics two articles that it returns. This is slightly better than random.(related article names are obtained from wikipedia API).
Right now, I'm maxed on daily Groq usage and the app falls back to pure random choice.
Why two articles? I was kind of going for a hot or not style system.
It just feels like browsing regular wikipedia but with two windows open. Either way, I forgot how much I love diving into knowledge rabbit holes and now have 5 different manifestos printed from the Slow movement (culture) wiki.
This is cool. I've made a browser extension for browsing the web in a more guided way and my first use case was a wikipedia tour guide. You type in a topic in the extension popup and it opens the side panel to track the journey, marking visited links and has a progress bar. it uses tab groups to keep the tour self contained and when you visit all the links it prompts you to close all the tabs.
The categories seem barely helpful at times. For biology I got food and microbiota, which is fair.
But then for Chemistry I got quantum mechanics and arabic language, for Metaphysics I got Newton's law of gravitation and pop culture.
Feels like some links are too tenuous to be useful, or the hallucination issue strikes again.
This is a good example of where AI will not take over.
Wikipedia rabbitholes do not follow any pattern. To truly create an AI that guides you through Wikipedia rabbithole you'd have to study thousands of humans going into actual rabbitholes, their clicks, and their reading patterns.
Otherwise it's just "let this AI take you on a journey you're not at all interested in".
OK, developer here. Here is how it works. It connects to groq and asks this
"I was going in this rabbit hole of wikipedia articles: #{chain} \n I have now found these topics:\n\n#{related_topics}\n\n Select the 2 most mind blowing topics from the list and that I'd enjoy given my rabbit hole. Ensure two topics are not the same."
and pics two articles that it returns. This is slightly better than random.(related article names are obtained from wikipedia API).
Right now, I'm maxed on daily Groq usage and the app falls back to pure random choice.
Why two articles? I was kind of going for a hot or not style system.
18 comments
[ 3.1 ms ] story [ 46.2 ms ] threadYou can also go to this page that happens to have links to every date page and you can clearly see that it's not picking anything related to the article contents as most of the time it just picks a random date rather than one of the related topics: https://wikidive.tulv.in/dive?topic=List_of_non-standard_dat...
I don't know why the creator doesn't just say that it's random instead of making an easily falsifiable claim that it's "AI-guided".
"I was going in this rabbit hole of wikipedia articles: #{chain} \n I have now found these topics:\n\n#{related_topics}\n\n Select the 2 most mind blowing topics from the list and that I'd enjoy given my rabbit hole. Ensure two topics are not the same."
and pics two articles that it returns. This is slightly better than random.(related article names are obtained from wikipedia API).
Right now, I'm maxed on daily Groq usage and the app falls back to pure random choice.
Why two articles? I was kind of going for a hot or not style system.
Anyway, it was a fun experiment. Learned lots.
Extension is available here:
https://chromewebstore.google.com/detail/unpack/mcgdbnjjnnfm...
But then for Chemistry I got quantum mechanics and arabic language, for Metaphysics I got Newton's law of gravitation and pop culture. Feels like some links are too tenuous to be useful, or the hallucination issue strikes again.
Wikipedia rabbitholes do not follow any pattern. To truly create an AI that guides you through Wikipedia rabbithole you'd have to study thousands of humans going into actual rabbitholes, their clicks, and their reading patterns.
Otherwise it's just "let this AI take you on a journey you're not at all interested in".
"I was going in this rabbit hole of wikipedia articles: #{chain} \n I have now found these topics:\n\n#{related_topics}\n\n Select the 2 most mind blowing topics from the list and that I'd enjoy given my rabbit hole. Ensure two topics are not the same."
and pics two articles that it returns. This is slightly better than random.(related article names are obtained from wikipedia API).
Right now, I'm maxed on daily Groq usage and the app falls back to pure random choice.
Why two articles? I was kind of going for a hot or not style system.
Anyway, it was a fun experiment. Learned lots.
What exactly is the "AI" needed for here?