Andrej Karpathy Explains His New YouTube Clone, HackerTube

What changed

Andrej Karpathy has launched HackerTube, a personal YouTube clone, as a direct response to his dissatisfaction with the current state of YouTube. He described YouTube as an "enshittified wall of slop" with "insanely exaggerated thumbnails, dumb recommendations of the same tired topics, truly terrible idiocracy-grade content." Karpathy built HackerTube over a Saturday morning using an AI model named Opus 5.5. The tool downloaded and filtered Hacker News comments that recommended YouTube videos, ultimately identifying 6,454 videos recommended between 2007 and 2026 across 150 topics. HackerTube features a YouTube-like UI, uses a 50% thumbnail size, offers custom recommendations without user accounts (storing state in local browser sessions), includes keyboard shortcuts, and allows data export for users to create their own versions. Karpathy stated the entire process took approximately 2 hours.

Why it matters for builders

Karpathy's project showcases a pragmatic approach to building a content platform tailored to a specific user's preferences, bypassing the perceived algorithmic failures of mainstream platforms. The use of AI for filtering and recommendation, combined with a focus on user control and data portability, offers a model for alternative platform development. It addresses the growing sentiment among some users that existing platforms are not serving their needs effectively.

Practical impact

Karpathy's HackerTube is a personal project, not a commercial venture, but it serves as a compelling example of how AI can be used to curate content based on community recommendations rather than engagement metrics alone. Builders concerned with content quality and user experience might find inspiration in its approach to recommendation systems and its emphasis on user agency. The project highlights a potential direction for niche content platforms that prioritize quality and user control.

Caveats and source limits

The HackerTube project is a personal endeavor by Andrej Karpathy and is presented as a solution for his own viewing preferences. The content quality and recommendation effectiveness are subjective and based on Karpathy's criteria and the filtering of Hacker News comments. The AI model Opus 5.5's specific capabilities and limitations in this context are not detailed. The project's scalability and long-term viability as a public platform are not addressed in the source.