HackerTube: YouTube built from videos recommended by HN

You can now find niche video recommendations based on curated community input, not just a broad algorithm.

Developer Andrej Karpathy expressed frustration with YouTube's recommendation algorithm, describing it as "enshittified" and filled with "slop." In response, he created HackerTube, a YouTube clone populated with videos recommended on Hacker News. Using an AI model called Opus 5.5, Karpathy processed approximately 40,000 comments from Hacker News dating back to 2007, identifying 6,454 YouTube video recommendations across 150 topics. The resulting platform, hackertube.org, features curated content, uses 50% thumbnails, offers personalized recommendations without user accounts, stores state locally, and includes keyboard shortcuts and data export capabilities. The entire process, from concept to deployment on a VPS, took Karpathy approximately two hours.

This project highlights a growing user dissatisfaction with algorithmic content curation on major platforms, pushing individuals to build bespoke solutions. By leveraging community-driven recommendations from a highly technical audience like Hacker News, Karpathy demonstrates a potential pathway for creating more relevant and quality-focused content discovery experiences. The rapid development time, aided by AI tools, suggests that personalized platforms can be created with significantly reduced effort.

The HackerTube model bypasses traditional user tracking and account-based personalization by relying on curated data and local browser state. This approach offers a privacy-conscious alternative and suggests a future where users have more control over their content feeds. The inclusion of data export further empowers users, enabling them to port their curated lists or build their own iterations, potentially fostering a more decentralized approach to content aggregation.