Show HN: Resursive URL/News site to learn the backstory on any article(getbackstory.ai)
getbackstory.ai
Show HN: Resursive URL/News site to learn the backstory on any article
https://getbackstory.ai
2 comments
I'd take just a 'scrape' of the text content of an article. Some sites take ridicilously long to load, don't work / display anything, or cause MBs of traffic for a few paragraphs of text (if that).
Separate that small % of content from the big % of overhead, for (almost) any site (news sites, in particular), and you have a winner.
Separate that small % of content from the big % of overhead, for (almost) any site (news sites, in particular), and you have a winner.
This was an issue I ran into a lot and why I settled eventually on a design similar to what you’re saying. Even for something seemingly simple like deciding what the title is or what’s the main picture is ridiculous hard with scraping alone so I have to pass most the data for ai analysis and then generate the details, summary and backstories. Thanks for checking it out
Nice idea. Hope it gets traction because it's sorely needed at the moment. It's not just about misinformation, it's all the horrendous clickbait.
Thanks, this is one of those - I need this to exist for me, and I’m hopeful others might care. I’m so tired of the “news” - just want to calm down, learn things, and see if society can find solutions.
Great idea. One suggestion might be to offer users a toggle between the square Recent Backstories cards and a much more condensed and easy to scan line by line river of news? Maybe even categorized? But great job anyway, so thanks.
Boulder's startup week has a hackathon going on right now, so this is my entry to try and make it happen. I'd been wanting to build a purely AI app for awhile, that doesn't support free form inputs (yes, I know...), and this one seemed intriguing with it's recursive design. It took me a little longer than I thought to get the articles right, but I think it works pretty well.
The first two paragraphs offer just a quick summary of the article (a user could link back too). But then, the next paragraph will explain the background concept, then the next will be more of a historical/scientific origin, then finally a deeper foundational level. It seems like the assumption should be that for the initially scanned URL, the user has probably already read it so no real need to rehash the whole thing. Then, for each story/article, it will generate 3 "Backstories". These will then be full deep educational articles that explain a full concept. Each of these 3 backstories will have a level too. So the first will have some immediate context to its parent story. The next will have a foundational perspective. The final will be origins/next steps. Then each of these "stories" will be recursive and allow a user to drill as deep as they want to go.
Scrapping all different formats of sites was trickier than I'd thought - Jina, puppeteer, perplexity, playwright, all couldn't quite get there especially with more complex views. The methods I ended up using were more involved so now I've had to introduce a credit system - we'll see how that goes.
Anyway, if anyone wants to check it out - let me know what you think. In my own testing, it seems fairly solid, but this was built pretty quickly.