4 ms·

Like a web of trust, but for link recommendations? Maybe it could work by changing the incentives. If you had a magical perfect AI that was loyal to you and ra

Like a web of trust, but for link recommendations? Maybe it could work by changing the incentives. If you had a magical perfect AI that was loyal to you and ran on your PC, it would always act in your interest - Spidering the web and returning interesting things based on your feedback. Whereas the current model is, none of this stuff is self-hosted, it has to run on Somebody Else's Computer. SEC costs money to run, so they have to make that money back by tweaking the recommendations, and none of the big ones inter-operate. YouTube will never recommend videos from the Fediverse. So anything that's not self-hosted (which is just a specific form of a subscription payment, anyway) will never have your best interests at heart, it can only have some of your interests at heart. This same problems comes up in politics - Companies have no interest in creating public goods and we shouldn't expect them to do so. But anyway, I think this is where some of the good stuff will fall out of AI in the next 10 or 20 years - Having intelligent agents that try to cut through the BS of the web on behalf of their owners.

frongpik 6y ago
My idea is a lot simpler and doesn't need any ML. Think of it as the linkedin feed: if you upvote a post, it'll be shown to your contacts, but if you upvote too much junk posts, your contacts will unsubscribe. This idea is relatively trivial to implement as an alternative UI of HN. That site would show the very same html, but will hack the upvote and downvote buttons. You upvote to subscribe to someone and downvote to unsubscribe. If someone in your watchlist upvotes something, that something gets shown in your feed. You'd still use the main HN feed for discovery. Eventually this would form a network that propagates information in some way. That network could be even visualised in some way to see how certain posts travel thru it. The topology of the network, its dynamic properties and whether it'll remain connected are interesting open questions. The entire thing can be written as a one file python script with a sqlite db running on a 5 bucks vps. The biggest challenge is getting enough folks on hn to use it.
- jaredsohn 6y ago
  I think there would need to be a browser extension as well to detect the upvotes/downvotes.
octoberfranklin 6y ago
> Like a web of trust, but for link recommendations? https://news.ycombinator.com/item?id=25405128 https://news.ycombinator.com/item?id=25405128

frongpik 6y ago

My idea is a lot simpler and doesn't need any ML. Think of it as the linkedin feed: if you upvote a post, it'll be shown to your contacts, but if you upvote too much junk posts, your contacts will unsubscribe. This idea is relatively trivial to implement as an alternative UI of HN. That site would show the very same html, but will hack the upvote and downvote buttons. You upvote to subscribe to someone and downvote to unsubscribe. If someone in your watchlist upvotes something, that something gets shown in your feed. You'd still use the main HN feed for discovery. Eventually this would form a network that propagates information in some way. That network could be even visualised in some way to see how certain posts travel thru it. The topology of the network, its dynamic properties and whether it'll remain connected are interesting open questions. The entire thing can be written as a one file python script with a sqlite db running on a 5 bucks vps. The biggest challenge is getting enough folks on hn to use it.

jaredsohn 6y ago
I think there would need to be a browser extension as well to detect the upvotes/downvotes.

octoberfranklin 6y ago

> Like a web of trust, but for link recommendations? https://news.ycombinator.com/item?id=25405128 https://news.ycombinator.com/item?id=25405128