A fresh Show HN submission is pitching a simple concept: Hacker News, but with AI-driven content pushed down the page. The project lives at hn-without-ai.vercel.app and launched on August 6, promising reduced priority for posts that look like they came out of an LLM rather than a human brain. The timing is no accident. HN has spent years wrestling with AI-generated submissions flooding the new queue โ low-effort blog spam, automated accounts posting 'insightful' comments at scale, and SEO bait dressed up as technical writing. Tools like this represent the DIY answer to that problem: instead of waiting for admins to fix ranking globally, just ship your own front-end that reorders things locally. Here's where it gets frustrating: there is almost no technical detail in the submission itself. At time of writing, the post has 2 points and zero comments, and nothing in the source material explains how the deprioritization actually works โ whether it's keyword-based filtering, a classifier model, or something simpler like flagging posts from known AI-content domains. Even as a bare proof of concept, though, the project surfaces an interesting design question: what does 'AI-driven content' mean in practice? A naive filter risks catching legitimate LLM-assisted writing while letting clever spam slip through. The real engineering challenge isn't detection โ it's precision and recall on a platform where plenty of humans now use AI tools to draft their work.
Key Takeaways
- New Show HN tool promises reduced priority for AI-generated posts, hosted at hn-without-ai.vercel.app.
- Launched August 6 with minimal traction so far: 2 points and no comments on the submission thread.
- No public documentation yet on how the ranking adjustment is implemented or tuned.
The Bottom Line
Worth a bookmark if you build community tools or care about content moderation. But with zero docs, no discussion, and almost no engagement, this reads as an idea sketch more than a shipped product โ whoever built it needs to show their work before anyone should trust the filter.