Curating Two Decades of Tech Discussions
On November 1, 2026, former Tesla AI Director Andrej Karpathy unveiled a weekend project titled HackerTube (hackertube.org), a platform curating 6,454 YouTube videos that have been most frequently cited and praised throughout Hacker News's 20-year history. Karpathy designed the project to counter the inefficiencies of YouTube's algorithmic feed by leveraging the collective knowledge and technical discussions of the developer community.
Built in Two Hours with Claude Opus 5.5
In a post on X, Karpathy revealed that he utilized Claude Opus 5.5 to ingest, parse, and filter the complete archive of 43 million comments accumulated over two decades on Hacker News. The entire pipeline—from data processing to building the web interface—took only about two hours. Karpathy emphasized that the viewing experience curated through this method delivers noticeably higher quality compared to YouTube's standard recommendation system.
A Ripple Effect Across Hacker News
Shortly after HackerTube went live, developers on Hacker News rapidly replicated the concept across various information formats. On the 'Show HN' forum, several spinoff projects emerged:
* HackerBooks (hackerbooks.org): Catalogs the most cited and debated books across the community. * Wikihack (wikihack.netlify.app): Aggregates and links Wikipedia entries referenced in forum threads. * Hacker Times (hackertimes.ca): Formats Hacker News articles in the classic editorial style of The New York Times.
This flurry of projects highlights an emerging trend of employing state-of-the-art AI models to transform massive historical forums into structured, high-signal curation engines.