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
Developer Andrej Karpathy says he built HackerTube, a YouTube-style site based on videos recommended in Hacker News comments. He reports that the project filters about 40,000 recommendation comments into a catalogue of 6,454 videos across 150 topics, but the source account does not independently verify the figures or explain how recommendations are ranked.
Andrej Karpathy says he has launched HackerTube, a YouTube-style site that gathers videos recommended in Hacker News comments, after finding YouTube’s recommendations poorly matched his tastes. The catalogue, as described in his October 31, 2026 blog post, contains 6,454 videos across 150 topics drawn from comments posted between 2007 and 2026; those figures come from Karpathy and have not been independently verified in the supplied material.
Karpathy said he started the project after opening YouTube late on a Friday night and finding its home page filled with recommendations he considered repetitive and low quality. The next morning, he asked an AI coding assistant, which he identified as Opus 5.5, to help build a YouTube clone using videos recommended on Hacker News. His description is a personal account of the motivation, not an independent evaluation of YouTube’s recommendation system or of the new site’s content quality.
According to Karpathy, the assistant processed Hacker News comments, filtered them, and used another model, which he called Sonnet, in a workflow intended to remove spam and low-quality material. He says the result was approximately 40,000 comments recommending YouTube videos, from which the project assembled its 6,454-video collection. He did not provide a detailed account of the filtering criteria, the models’ specific roles, or how duplicate links and repeated recommendations were handled.
Karpathy said the site presents the collection through a familiar video-browsing interface and includes topic categories, custom recommendations, keyboard shortcuts and an option to export the site’s data. He also says recommendations work without user accounts and that browsing state stays in the local browser session. The post describes the product’s intended design; it does not include an independent privacy review, technical audit or usage figures.
A Different Route to Video Discovery
HackerTube offers an example of recommendation based on a community’s past links rather than a platform’s personalized feed. For viewers dissatisfied with mainstream recommendations, a collection assembled from Hacker News discussions could provide another way to find talks, demonstrations and other videos that people chose to share in a technical community. That does not establish that the collection is broadly representative or consistently high quality: its selection reflects the interests and posting habits of Hacker News users.
The project also puts a spotlight on how AI tools can help turn a large archive of public discussion into a usable catalogue. Karpathy says the work—from processing comments to building and deploying the site—took about two hours. That is his estimate of this particular build, not a general measure of how quickly similar services can be made or maintained. The approach depends on the source data, filtering choices and the availability of working video links.
Its no-account design may appeal to people who prefer not to create a profile for recommendations. At the same time, the post does not explain whether the site records server-side access data, how its video embeds behave, or how its recommendations change as someone browses. Those details matter to readers evaluating privacy and whether the experience differs meaningfully from a conventional feed.
Products Worth Considering
Bose QuietComfort Ultra Wireless Noise-Cancelling Headphones with Spatial Audio, Black
Logitech MX Mechanical Wireless Illuminated Performance Keyboard – Clicky
Bose QuietComfort Wireless Noise Cancelling Headphones, Sandstone
As an affiliate, we earn on qualifying purchases.
From Hacker News Links to a Catalogue
Hacker News is a discussion forum where users submit stories and comment on them. Members sometimes share YouTube videos in discussions, but those links were not necessarily posted as formal recommendations or with consistent descriptions. HackerTube’s premise is to treat those scattered mentions as a source for a video library and present them in a dedicated interface.
Karpathy’s account describes a chain of processing: downloading comments, filtering them, applying an AI-assisted workflow to screen the results, and extracting a set of video recommendations. He says the source period runs from 2007 through 2026, giving the collection a long historical span. The post does not state the precise date on which the comment archive was captured, so the endpoint should be understood as the period he reports rather than a verified cutoff.
Karpathy also says HackerTube uses a video’s 50% thumbnail rather than its uploaded thumbnail. He presents this, along with the interface and recommendations, as a design feature. The supplied account does not explain how that thumbnail is selected or provide comparative evidence that it improves discovery.
How the Collection Was Filtered
The supplied material is Karpathy’s own post, so the reported comment count, video total and topic count have not been independently checked here. The post does not link to a methodology explaining what counted as a recommendation, how spam or low-grade content was defined, or how the AI-assisted screening was reviewed. It also does not say whether the catalogue includes every eligible video link or only those that passed the project’s filters.
Other practical details remain unspecified. The account gives no traffic or usage numbers, no information on how often the collection is refreshed, and no explanation of how broken, removed or private YouTube videos are handled. It does not describe the recommendation logic behind the site’s “ForYou” feature, beyond saying it is custom and works without accounts or stored state outside the local browser session.
The post also does not provide an independent security or privacy assessment, despite saying the service was secured and hardened before deployment. The extent of any server-side logging, the site’s ongoing operating arrangements and the models’ exact contributions are not clear from the source.
Checking the Site’s Ongoing Operation
Karpathy’s post presents HackerTube as a deployed site, but it does not announce a release schedule, maintenance plan or next product milestone. The immediate next step for readers is to consult the live site and its available data export, while keeping in mind that the source account alone does not verify the catalogue or its methods. Whether the project will add new Hacker News recommendations, correct unavailable links or publish more details about its filtering process remains unknown.
Further information from the creator about collection dates, recommendation rules, privacy practices and ongoing updates would help clarify how the service works beyond the initial description. No such follow-up or independent assessment is included in the supplied material.
Key Questions
What is HackerTube?
HackerTube is a YouTube-style browsing site that Karpathy says is built from video links recommended in Hacker News comments. He describes a collection of 6,454 videos across 150 topics.
Where did HackerTube’s videos come from?
Karpathy says the project processed Hacker News comments from 2007 to 2026, identifying about 40,000 comments that recommended YouTube videos before filtering the material into the catalogue. The exact capture date and filtering method are not specified.
Does HackerTube require an account?
Karpathy says the site’s recommendations work without user accounts and that browsing state is kept in the local browser session. His post does not provide an independent privacy review or details on server-side logging.
How are videos recommended on HackerTube?
The site is based on videos shared in Hacker News comments, and Karpathy says it offers custom “ForYou” recommendations. The post does not explain the recommendation algorithm or how it responds to a visitor’s browsing.
Are HackerTube’s figures independently verified?
Not in the supplied source material. The totals of 40,000 comments and 6,454 videos are figures reported by Karpathy in his blog post, with no independent audit or detailed methodology provided there.
Source: karpathy.bearblog.dev
