I wanted to make this general purpose enough that it could handle large datasets (e.g. i want to do thiings like https://sprinklz.io/demo/headlineNews), so I was hesitant to do things like make 1-LLM-call-per-article, but I'll keep exploring ways to make it better
When I load this page, it shows DeepMind CEO's departure as the #1 post, and DeepMind to me seems like an AI driven organization. Qwen shows up at #6, which is an AI model.
it basically does a combination of:
-- string text matching + embeddings
unfortunately there isn't a dashboard yet in the logged out view, but if you view the feed when logged into sprinklz you can directly see (and edit) all the params used
At its core, this isn’t merely a tool that navigates the complexities of the digital landscape—it’s a load-bearing solution that deliberately removes content generated using AI, rather than content about LLM models.
It’s worth noting that we’re not delving into the rich tapestry of discussions surrounding large language models themselves; instead, we’re underscoring the pivotal distinction between AI-generated material and authentic human writing about those very models. Ultimately, this approach fosters a more robust, seamless, and nuanced filtering process—one that leverages genuine insight rather than the multifaceted outputs of generative systems.
The emdashes from your LLM-generated responses are leaking. I would encourage you to consider writing comments here as your own, which better reflects you as a person.
it basically does a combination of:
-- string text matching + embeddings
unfortunately there isn't a dashboard yet in the logged out view, but if you view the feed when logged into sprinklz you can directly see (and edit) all the params used
Seems to work - "Macron's 'Free speech is pure bullshit.' and the hand that hollows free speech" - an AI generated (and not very well edited) ad for a product called "Veri" is on the front page of HN but much lower on this site.
Love this idea. I agree with the other comment that its not filtering well, but i hope you can get it dialed in.
Sometimes hackernews feels like the drive into SF. Just bombarded with Ai garbage billboards.
nice, yeah, I will keep working on it
I wanted to make this general purpose enough that it could handle large datasets (e.g. i want to do thiings like https://sprinklz.io/demo/headlineNews), so I was hesitant to do things like make 1-LLM-call-per-article, but I'll keep exploring ways to make it better
What is "AI driven content"?
When I load this page, it shows DeepMind CEO's departure as the #1 post, and DeepMind to me seems like an AI driven organization. Qwen shows up at #6, which is an AI model.
Something seems deeply flawed about this website.
it basically does a combination of: -- string text matching + embeddings
unfortunately there isn't a dashboard yet in the logged out view, but if you view the feed when logged into sprinklz you can directly see (and edit) all the params used
At its core, this isn’t merely a tool that navigates the complexities of the digital landscape—it’s a load-bearing solution that deliberately removes content generated using AI, rather than content about LLM models.
It’s worth noting that we’re not delving into the rich tapestry of discussions surrounding large language models themselves; instead, we’re underscoring the pivotal distinction between AI-generated material and authentic human writing about those very models. Ultimately, this approach fosters a more robust, seamless, and nuanced filtering process—one that leverages genuine insight rather than the multifaceted outputs of generative systems.
The emdashes from your LLM-generated responses are leaking. I would encourage you to consider writing comments here as your own, which better reflects you as a person.
itsajoke.fbx
Why a JavaScript loading screen? HN works without JS.
it isn't visible from this view, but it's designed so that you can edit the ranking params directly from your browser, e.g. https://sprinklz.io/demo/headlineNews or https://sprinklz.io/demo/codeEditor
You're still reading code like it's 2020? Wow
Article about Qwen was at sixth place when I loaded the page, followed by gpt-5.6 sol.
I guess this is still an improvement, today’s front page often contains multiple AI adjacent submissions in the top ten.
How is this curated? Don't see an explanation on the site.
it basically does a combination of: -- string text matching + embeddings
unfortunately there isn't a dashboard yet in the logged out view, but if you view the feed when logged into sprinklz you can directly see (and edit) all the params used
I guess this was vibe-coded? The top three item are "AI"-driven content. And obviously so since each says AI in the very first para.
this is not vibe coded.
it is part of a much larger project than just this page (https://sprinklz.io), which I have been working on, on and off, for almost 4 years.
Seems to work - "Macron's 'Free speech is pure bullshit.' and the hand that hollows free speech" - an AI generated (and not very well edited) ad for a product called "Veri" is on the front page of HN but much lower on this site.