TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
HackerTube is a new video-browsing site assembled from YouTube links recommended in Hacker News comments. Its creator says an automated process filtered the comments to identify 6,454 videos across 150 topics, with recommendations that do not require user accounts. The filtering method and the quality of the resulting collection have not been independently evaluated.
Developer Andrej Karpathy has published HackerTube, a YouTube-style site that organizes videos recommended in Hacker News comments. Karpathy says the collection contains 6,454 videos drawn from recommendations posted between 2007 and 2026, offering users an alternative way to browse YouTube without relying on their personal account history.
In a post dated October 31, 2026, Karpathy described building the site after finding his YouTube recommendations unhelpful. He says he tasked an AI coding system, which he calls Opus 5.5, with creating a video site based on links mentioned on Hacker News. The post says the project took about two hours; that timeframe is Karpathy’s account, not an independently verified measure.
According to the post, the process downloaded Hacker News comments, filtered them, and used a Sonnet model in a workflow to remove spam and what Karpathy described as low-quality material. The result was a set of about 40,000 comments that recommended YouTube videos. Those comments pointed to 6,454 videos spanning 150 topics, according to the creator.
HackerTube presents the links in a familiar YouTube-style interface. Karpathy says it uses a cropped thumbnail rather than the original uploaded thumbnail and provides custom “For You” recommendations without user accounts. The post also describes keyboard shortcuts and an option to export the site’s data for use in creating another version. Details of the recommendation logic and the available topic categories were not provided in the source account.
A Different Route to Video Discovery
HackerTube tests a simple alternative to a feed built around an individual viewer’s watch history and account data: selecting videos that people have linked in a public discussion forum. That changes the source of recommendations, but does not by itself establish that the resulting feed is more useful, more accurate, or higher quality for a broad audience.
The project also makes a large set of past community recommendations easier to browse than a comment archive. The collection could help people find videos that received attention in Hacker News discussions, while its historical span makes it possible to encounter older links alongside newer ones. The site’s selection still reflects who participates on Hacker News and what they choose to post, rather than a representative sample of YouTube viewers or videos.
Its account-free approach may appeal to people who want recommendations without signing in or adding viewing activity to a personal profile. Karpathy says the site keeps state in the local browser session, but the post does not provide a technical audit of data handling. The project is best understood as a creator-built experiment in recommendation and interface design, not as evidence that one recommendation method outperforms another.
Products Worth Considering
Bose QuietComfort Wireless Noise Cancelling Headphones, Sandstone
Bose QuietComfort Ultra Wireless Noise-Cancelling Headphones with Spatial Audio, Black
Bose QuietComfort Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Black
As an affiliate, we earn on qualifying purchases.
From Forum Links to a Video Site
Hacker News is a technology-focused discussion site where users post links and comment on them. Videos can appear in those discussions as recommendations, examples, or material related to a story. HackerTube takes those scattered links and presents them in a dedicated video-browsing interface.
Karpathy’s post describes the development as a response to dissatisfaction with his own YouTube recommendations. He characterized the feed he saw as repetitive and dominated by exaggerated thumbnails, but that is his personal assessment, not a finding about YouTube’s recommendations for users generally. He then used an AI-assisted workflow to collect and filter the comments, before putting the resulting links into a site hosted on a virtual private server, according to his account.
The stated collection covers recommendations from 2007 through 2026. That date range indicates the period represented in the comments, not necessarily when every linked video was published. The post does not describe a separate review of the videos themselves or compare HackerTube’s recommendations with YouTube’s own feed.
Filtering and Feed Details Remain Open
The available account is Karpathy’s own description of the project. The source material does not include an independent review of the comment-filtering process, the accuracy of the extracted links, or the criteria used to decide which comments counted as recommendations. It is also unclear how many comments were excluded, how duplicates or broken links were handled, and whether the collection will be updated.
The post does not explain how HackerTube ranks or selects videos for its “For You” feed, beyond saying that its recommendations are custom and do not require an account. It gives no performance data, user numbers, comparison with YouTube’s recommendation system, or evaluation of whether viewers find the selections useful. The claims about data remaining in a local browser session and about the site’s security are likewise not independently verified in the source material.
It is not clear whether the site has formal permission or arrangements for using YouTube’s video links and thumbnails, whether videos can be removed from the collection, or how the export handles changes to the source data. The announcement also does not establish how the site will be maintained or whether its data will remain available.
Site Use and Maintenance Ahead
Karpathy’s post presents HackerTube as a working site and says its data can be exported, but it does not announce a release schedule, maintenance plan, or future feature roadmap. The next useful developments would be details from the creator about how the feed is generated, whether the source collection will be refreshed, and how the export can be used.
Further reporting could also check the site’s current availability and inspect how its local-session claim works. Until there is an independent review or more information from the creator, the figures describing the collection and the account of the build process should be treated as claims in the original announcement, rather than independently confirmed results.
Key Questions
What is HackerTube?
HackerTube is a YouTube-style site that organizes videos linked in Hacker News comments, according to its creator, Andrej Karpathy.
How many videos does HackerTube include?
Karpathy says the site includes 6,454 videos identified from about 40,000 comments, covering recommendations posted from 2007 to 2026 across 150 topics. Those figures come from his announcement and have not been independently checked in the supplied source material.
Does HackerTube require a user account?
Karpathy says its custom “For You” recommendations work without user accounts and that state is kept in the local browser session. The source account does not include an independent technical review of how the site handles data.
How were the video links selected?
According to Karpathy, the project collected Hacker News comments, filtered them, and used a Sonnet model in a workflow to remove spam and low-quality material. The announcement does not give the full filtering rules or describe an independent assessment of the selected links.
Is HackerTube proven to give better recommendations than YouTube?
No such comparison is reported in the source material. HackerTube uses a different source for its video collection—recommendations in Hacker News comments—but there is no user study or performance data showing that its selections are better for viewers.
Source: karpathy.bearblog.dev
