Karpathy Launches HackerTube Video Site
The site collects 6,454 videos recommended in Hacker News comments and sorted into 150 topics.
Updated on Nov. 1, 2026 in Artificial Intelligence
Live Poll
Would you use a video site curated by AI from online community recommendations?
Andrej Karpathy launched HackerTube on October 31, 2026, a video site at hackertube.org containing 6,454 videos recommended in Hacker News comments. The videos are sorted into 150 topics, and Anthropic's Claude models were used to build the project.
Why it matters
The project shows AI models being used to curate community-sourced content at scale, turning years of informal recommendations into a searchable video collection. Its recommendation system runs without user accounts, storing data in each user's browser.
The build consumed 504 million input tokens, 90% of which were cached, and generated 7.6 million output tokens. Accuracy of the filtering against human judgment has not been reported.
The players
Andrej Karpathy
A prominent AI researcher and former Tesla AI director who initiated the HackerTube project.
Anthropic
The AI company whose Claude models were used to manage and filter the HackerTube project.
The details
Claude Opus 5.5 managed the project while Claude Sonnet performed the filtering, processing nearly 40,000 comments in about two hours. A video needed recommendations from multiple users across different months to qualify, and Karpathy's own role was limited to obtaining a Google Cloud API key, spinning up a VPS, and buying the domain.
Timeline
- 2007: Earliest Hacker News comments analyzed for HackerTube.
- October 31, 2026: HackerTube launched.
The Tech Race
The project illustrates how frontier models from Anthropic are being applied beyond chat and coding to autonomous, multi-step data curation tasks. It positions Claude as a tool for building complete small products rather than only assisting developers.
Users can browse the 6,454 videos at hackertube.org without creating an account, since recommendations are stored in the browser. The site offers a searchable library of community-vetted technical videos organized into 150 topics.
The takeaway
Community discussion can be turned into a structured resource with relatively little manual effort when AI models handle the filtering. Readers interested in similar projects can review how the comments were selected and ranked.
Further reading
For more on the AI tools behind projects like this, see Artificial Intelligence.
More information
Browse the videos on the HackerTube video recommendation site.
Live Poll
Would you use a video site curated by AI from online community recommendations?
