Andrej Karpathy launches HackerTube, a Claude-curated YouTube alternative built from Hacker News comments
Former Tesla AI chief Andrej Karpathy has launched HackerTube, a video platform featuring 6,454 community-vetted videos selected from Hacker News comments using Anthropic's AI models.
Launched on October 31, 2026, at hackertube.org, the platform represents a departure from traditional algorithmic content recommendation. Instead, it relies on the collective judgment of Hacker News commenters, a community known for its technical expertise and strong opinions. The project processed nearly 40,000 comments spanning from 2007 to 2026, requiring videos to have multiple recommendations across different months to qualify.
AI-powered curation
Karpathy leveraged Anthropic's AI models to build the platform efficiently. Claude Opus 5.5 managed the project while Claude Sonnet handled the filtering work. The system consumed approximately 504M input tokens (with 90% cached) and 7.6M output tokens during the two-hour processing period. This approach allowed Karpathy to personally handle only three tasks: obtaining a Google Cloud API key, setting up a VPS, and purchasing the domain name.
Platform features
HackerTube distinguishes itself with several user-friendly features. Thumbnails use a 50% video-frame treatment, and the platform supports keyboard shortcuts for navigation. Recommendations are stored locally in users' browsers rather than requiring accounts, enhancing privacy. Users can also download the entire dataset as a zip file for customization.
The bar for inclusion was not a single enthusiastic post. A video needed recommendations from multiple users, and those recommendations had to come across different months.
Community-driven approach
The platform's curation philosophy diverges from mainstream content recommendation by emphasizing community validation over algorithmic selection. This method filters out one-time hype and prioritizes content with staying power. The resulting 150 topics reflect the interests of the Hacker News community, which skews toward technical and engineering content.
For Anthropic, the project serves as an unpaid showcase of their models' capabilities, demonstrating a practical workflow where Opus handles coordination while Sonnet manages repetitive tasks. The project also highlights the efficiency gains from token caching in large-scale AI processing.
Related
