Show HN: Graph – turn your ChatGPT into AI-sorted RSS feeds

graph.cx

artificial intelligenceRSS feedspersonalization

Discussion (6 comments)

Showing 6 comments

1zael

4 months ago

1 reply

This is clever...using embeddings to turn RSS from a source-following model into a topic-following one based on your daily LLM usage feels like the straightforward evolution of RSS. Have you considered adding a "serendipity slider" to occasionally inject high-quality off-topic content? One of RSS's hidden benefits was stumbling onto things outside your filter bubble.

jbg100

4 months ago

hey, thank you! we have also thought about that! in "find" aka search, you'll see a button called "tumble" which just throws in a randomized search word, to help stumble upon something you wouldn't have otherwise.

ha I went into a rabbit hole on IMF yesterday, having not known a lot about it going in. thanks for checking this out!

samjgorman

4 months ago

1 reply

Nice work! To share some thoughts on the UX of Graph...

1. The visual design of the app feels largely different from other apps that I use, which is largely good. It's evoking an early internet aesthetic. To what degree were early social communities like MySpace an inspiration? 2. The other side of breaking from more modern UX patterns is that users might stumble or get slowed down. I had a harder time figuring out the UX of the chat inbox, likely because I was bringing iMessage style mental models to more of an AIM UX. 3. Onboarding of inputting ChatGPT derived preferences to quickly personalized was really smart and can imagine this could be the start of a new way to shortcut otherwise tedious onboarding preference collection 4. Did you consider making this an iOS app? I almost only read HN on my phone these days and would likely appreciate these more niche articles tailored towards learning compared to a standard news digest

jbg100

4 months ago

Great questions, ty!

1. The UI (for now at least) is definitely inspired by olds. Some have called it MySpace x StumbleUpon. 2. Agree going retro future has its pros and cons tho and the point about chat UI path dependency. We were in fact leaning more AIM/MSN Messenger w that, and are hoping it clicks more when we roll out Graph chat soon. 3. Glad the onboarding 0 - 100 clicked. Hope that saves time for folks. Fun fact is I wanted to build something like this years ago, but didn't see ppl having the time to hand type dozens, much less, 100 key interests or media likes. 4. For sure would love to make this a native mobile app. It's a hack but I've gotten far in using the iOS shortcut UX to make a shortcut app of it on my phone. Maybe this can be a holdover for some before we get opt-in email digests live.

Thanks for the great questions and checking it out!

crystal_clear

4 months ago

1 reply

Congrats on the launch! How is the homepage feed right now sorted? Is it by chronological order or is there a different type of logic (eg the topmost one is the most likely to be engaging)?

jbg100

4 months ago

thanks! it's the latter. each post shown to a user hit the internet in the last 0-24 hours and has a signal rating between 0 and 1. so the ones are the top are more likely to match. evolving this soon, but for now we have a "super tag" feature that can be used to boost topics you like most. Thanks for checking things out!

Resources