TL;DR
Get everyday helpers delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
A developer known as karpathy says an AI-assisted project called HackerTube turns YouTube videos recommended in Hacker News comments into a searchable, YouTube-style site. The collection contains 6,454 videos across 150 topics, drawn from 40,000 selected comments; the project’s creator says the work took about two hours. The site’s recommendation quality and its longer-term operation have not been independently assessed in the source material.
Developer karpathy has published HackerTube, a YouTube-style site built around videos recommended in Hacker News comments, saying an AI-assisted workflow gathered 6,454 videos from recommendations posted between 2007 and 2026. The project offers a different way to browse those links, without requiring a user account, and the creator says building it took about two hours.
In a post dated October 31, 2026, karpathy described creating the site after finding YouTube’s recommendations unsuited to their tastes. They said they asked an AI model identified as Opus 5.5 to help make a video site based on videos recommended on Hacker News. The report says the resulting project is available at hackertube.org.
According to the creator’s account, the process began with downloading Hacker News comments and filtering them for YouTube recommendations. A workflow using Sonnet then helped remove spam and lower-quality material, leaving about 40,000 comments that recommended a video. Those comments yielded 6,454 videos, spanning 150 topics and recommendations made from 2007 through 2026.
Karpathy said HackerTube presents the collection in a familiar YouTube-style interface and uses thumbnail images taken from the videos rather than the uploaded thumbnails. The post also describes custom recommendations, keyboard shortcuts, and the ability to export the site’s data for use in making another version. The creator says the site stores browsing state in the local browser session and does not require accounts.
A Community-Sourced Alternative Feed
HackerTube’s central idea is to use links recommended by a community as the starting point for video discovery, instead of relying only on the platform’s own recommendation feed. That makes the project a practical experiment in how a video library can be assembled from public discussion and presented through a familiar interface.
The size and age of the collection give visitors access to recommendations posted across nearly two decades, rather than only links circulating today. The project also puts data portability in view: its creator says visitors can export the site’s data and use it to build a version of their own. The source does not establish whether the collection or interface leads to better recommendations for users generally.
Popular right now
TCL QM64L 85-Inch QD-Mini LED 4K Smart Fire TV
Bose QuietComfort Wireless Noise Cancelling Headphones, Sandstone
Dyson Cyclone V10 Origin Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
From HN Comments to Videos
Hacker News is a discussion site where users share links and comment on technology and other subjects. HackerTube draws on video links that users recommended in those discussions, rather than presenting itself as a new video-hosting service. The videos remain YouTube videos; the project provides a separate way to find and browse selected links.
The creator’s account describes a pipeline that filters comments, screens recommendations for spam and quality, and groups the resulting videos across 150 topics. The original post characterizes the final set as recommendations made by “a few people on HN,” but does not provide a detailed explanation of how often recommendations appeared or how the topic categories were assigned.
“Let’s write our own YouTube clone based on whatever videos have been recommended on Hacker News.”
— Karpathy, in the HackerTube project post
Questions About the Selection Process
The figures and descriptions of the filtering process come from the project creator’s post; the supplied report does not include an independent audit of the comment set, video count or classification. It does not specify how the system defined a high-quality recommendation, how many comments were excluded, or whether multiple recommendations of the same video were counted or combined.
The source also gives no information about the site’s ongoing maintenance, traffic, or the availability of every linked video. It is not clear how HackerTube’s custom recommendations work in detail, or whether they use only the collected comments and topic labels. No user testing or comparison with YouTube’s recommendations is reported.
The Site’s Next Steps
HackerTube is presented as a live project at hackertube.org, with data export offered as a way for others to create their own version. The source material does not announce a release schedule, planned features, or a maintenance roadmap. Its next developments will depend on whether the creator updates the collection and whether other users make use of the site or its exported data.
For now, the published description establishes the project’s stated scope and features, but not how its recommendations perform in practice. Further information about its filtering rules, data updates and continued operation would clarify how the collection changes over time.
Key Questions
What is HackerTube?
HackerTube is a YouTube-style browsing site built around videos recommended in Hacker News comments. It links to YouTube videos rather than hosting them, according to the project description.
How many videos does the site include?
Karpathy says the collection has 6,454 videos across 150 topics, selected from about 40,000 comments recommending YouTube videos.
What years do the recommendations cover?
The creator says the recommendations were posted on Hacker News between 2007 and 2026.
Does HackerTube require an account?
The project post says the site works without user accounts and stores state in the local browser session. The source does not provide further technical details about data handling.
Has the site’s recommendation quality been independently tested?
The supplied report describes the creator’s filtering and recommendation features but provides no independent evaluation or user-testing results.
Source: karpathy.bearblog.dev
