# HackerTube: Karpathy builds a YouTube clone from Hacker News recommendations [Canonical reading](https://escapemeridian.com/readings/hackertube-karpathy-builds-a-youtube-clone-from-hacker-news-35af1598) Andrej Karpathy says he turned to YouTube on a Friday night and found its recommendations full of “the same tired topics.” The next morning, he asked Opus 5.5 to build a YouTube clone using videos recommended on Hacker News. He says the process gathered and filtered Hacker News comments, leaving 40,000 comments recommending 6,454 videos from 2007–2026 across 150 topics, then put the videos in a YouTube-style interface. He describes HackerTube as offering custom recommendations without accounts, using local browser session state, and letting users export the site’s data to make their own version. Karpathy says the work took about two hours, but his account also names tasks he handled: creating a Google Cloud API key, starting a VPS, and buying a domain. In an edit, he says he no longer does traditional engineering such as reading or writing code or running terminal commands, and instead tells Claude to do the work. His description presents the project as a personal alternative to YouTube’s recommendations and as an example of working with Claude; it does not report how users responded or how the recommendations performed. He calls the result “quality content,” but does not explain how that quality was assessed. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source Andrej Karpathy · Article Original source date: Oct 31, 2026 (2026-10-31T17:00Z). Date basis: Date stated by the publisher. Added to Escape Meridian: 2026-11-11T03:23:06.273Z. First observed: 2026-10-31T19:09:44.987Z. Retrieved: 2026-11-01T03:01:06.193Z. - [HackerTube](https://karpathy.bearblog.dev/hacker-tube/) ## Limits The account is Karpathy’s description of his own project; it does not report independent evaluation of the video selection or recommendation quality. The post does not describe how the videos were selected from the comments beyond filtering and removing spam, or how HackerTube’s recommendations perform for users. ## Open question How does HackerTube determine which recommended videos count as “quality content”? ## Source-stated Karpathy says he asked Opus 5.5 to make a YouTube clone based on videos recommended on Hacker News. > Let's write our own YouTube clone based on whatever videos have been recommended on Hacker News. - [Supporting source](https://karpathy.bearblog.dev/hacker-tube/) ## Source-stated He says the process narrowed Hacker News comments to 40,000 recommendations for 6,454 YouTube videos across 150 topics. > narrowed everything down to 40K comments recommending a YouTube video. In total, we get 6,454 YouTube videos - [Supporting source](https://karpathy.bearblog.dev/hacker-tube/) ## Source-stated Karpathy describes HackerTube as offering recommendations without user accounts, with state stored in the local browser session. > Custom and good ForYou recommendations without any user accounts, storing state in local browser session - [Supporting source](https://karpathy.bearblog.dev/hacker-tube/) ## Source-stated He says his own work involved creating a Google Cloud API key, starting a VPS, and buying a domain, while Claude handled the code and deployment tasks. > the only "manual work" I did was: 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box and 3) I bought the domain name. - [Supporting source](https://karpathy.bearblog.dev/hacker-tube/) ## Escape Meridian interpretation Karpathy calls the result “quality content,” but gives no account in this post of how quality was evaluated. > HackerTube: > Contains quality content - [Supporting source](https://karpathy.bearblog.dev/hacker-tube/) ## Source and contributor context Andrej Karpathy · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media declared by the original publisher; availability checked separately from the editorial reading. - [image · Andrej Karpathy](https://bear-images.sfo2.cdn.digitaloceanspaces.com/karpathy/image-2.webp) ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## API An application programming interface: a defined way for software to request data or actions from another system. - [Definition reference](https://developer.mozilla.org/en-US/docs/Glossary/API) ## Andrej Karpathy Concrete implementations and explanations of machine learning; useful teaching artifacts. - [Explore this context](/entities/andrej-karpathy) - [Reference](https://www.dwarkesh.com/p/andrej-karpathy) - [Reference](https://lexfridman.com/andrej-karpathy) - [Reference](https://www.dwarkesh.com/p/hiring-scouts-to-help-me-find-guests) - [Reference](https://karpathy.ai/) - [Reference](https://karpathy.github.io/) ## Claude Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding. - [Explore this context](/entities/claude) - [Reference](https://claude.com/product/overview) - [Reference](https://www.anthropic.com/news) - [Reference](https://www.anthropic.com/research/team/economics) ## Editorial provenance Version 1. Publication ID: bcbd4f6e-eeb3-4064-87bc-d1a9e4295dfd. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T03:23:06.273Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: useful\_perspective. Observation: 33f1964d8ba88c6fbe44b08c1a21c9fe2a4f23ee4b1bf8a30151317ef70b58fc. Source text hash: ae668dba51220bf2f60e22679b8bea41ab192a7b6ff546b117a2684a9047b926. ## Revision and source-check history ### Version 1 — 2026-11-11T03:23:06.273Z Andrej Karpathy says he turned to YouTube on a Friday night and found its recommendations full of “the same tired topics.” The next morning, he asked Opus 5.5 to build a YouTube clone using videos recommended on Hacker News. He says the process gathered and filtered Hacker News comments, leaving 40,000 comments recommending 6,454 videos from 2007–2026 across 150 topics, then put the videos in a YouTube-style interface. He describes HackerTube as offering custom recommendations without accounts, using local browser session state, and letting users export the site’s data to make their own version. Karpathy says the work took about two hours, but his account also names tasks he handled: creating a Google Cloud API key, starting a VPS, and buying a domain. In an edit, he says he no longer does traditional engineering such as reading or writing code or running terminal commands, and instead tells Claude to do the work. His description presents the project as a personal alternative to YouTube’s recommendations and as an example of working with Claude; it does not report how users responded or how the recommendations performed. He calls the result “quality content,” but does not explain how that quality was assessed. ## Original source Andrej Karpathy · Article Original source date: Oct 31, 2026 (2026-10-31T17:00Z). Date basis: Date stated by the publisher. Added to Escape Meridian: 2026-11-11T03:23:06.273Z. First observed: 2026-10-31T19:09:44.987Z. Retrieved: 2026-11-01T03:01:06.193Z. - [HackerTube](https://karpathy.bearblog.dev/hacker-tube/) ## Limits The account is Karpathy’s description of his own project; it does not report independent evaluation of the video selection or recommendation quality. The post does not describe how the videos were selected from the comments beyond filtering and removing spam, or how HackerTube’s recommendations perform for users. ## Open question How does HackerTube determine which recommended videos count as “quality content”? ## Source-stated Karpathy says he asked Opus 5.5 to make a YouTube clone based on videos recommended on Hacker News. > Let's write our own YouTube clone based on whatever videos have been recommended on Hacker News. - [Supporting source](https://karpathy.bearblog.dev/hacker-tube/) ## Source-stated He says the process narrowed Hacker News comments to 40,000 recommendations for 6,454 YouTube videos across 150 topics. > narrowed everything down to 40K comments recommending a YouTube video. In total, we get 6,454 YouTube videos - [Supporting source](https://karpathy.bearblog.dev/hacker-tube/) ## Source-stated Karpathy describes HackerTube as offering recommendations without user accounts, with state stored in the local browser session. > Custom and good ForYou recommendations without any user accounts, storing state in local browser session - [Supporting source](https://karpathy.bearblog.dev/hacker-tube/) ## Source-stated He says his own work involved creating a Google Cloud API key, starting a VPS, and buying a domain, while Claude handled the code and deployment tasks. > the only "manual work" I did was: 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box and 3) I bought the domain name. - [Supporting source](https://karpathy.bearblog.dev/hacker-tube/) ## Escape Meridian interpretation Karpathy calls the result “quality content,” but gives no account in this post of how quality was evaluated. > HackerTube: > Contains quality content - [Supporting source](https://karpathy.bearblog.dev/hacker-tube/) ## Source and contributor context Andrej Karpathy · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media declared by the original publisher; availability checked separately from the editorial reading. - [image · Andrej Karpathy](https://bear-images.sfo2.cdn.digitaloceanspaces.com/karpathy/image-2.webp) ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## API An application programming interface: a defined way for software to request data or actions from another system. - [Definition reference](https://developer.mozilla.org/en-US/docs/Glossary/API) ## Andrej Karpathy Concrete implementations and explanations of machine learning; useful teaching artifacts. - [Explore this context](/entities/andrej-karpathy) - [Reference](https://www.dwarkesh.com/p/andrej-karpathy) - [Reference](https://lexfridman.com/andrej-karpathy) - [Reference](https://www.dwarkesh.com/p/hiring-scouts-to-help-me-find-guests) - [Reference](https://karpathy.ai/) - [Reference](https://karpathy.github.io/) ## Claude Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding. - [Explore this context](/entities/claude) - [Reference](https://claude.com/product/overview) - [Reference](https://www.anthropic.com/news) - [Reference](https://www.anthropic.com/research/team/economics) ## Editorial provenance Version 1. Publication ID: bcbd4f6e-eeb3-4064-87bc-d1a9e4295dfd. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T03:23:06.273Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: useful\_perspective. Observation: 33f1964d8ba88c6fbe44b08c1a21c9fe2a4f23ee4b1bf8a30151317ef70b58fc. Source text hash: ae668dba51220bf2f60e22679b8bea41ab192a7b6ff546b117a2684a9047b926. ### uncertain — 2026-11-11T03:23:06.273Z — anthropic/claude-opus-5.5 Publication: bcbd4f6e-eeb3-4064-87bc-d1a9e4295dfd attribution — pass: The reading attributes the project, process and judgments to Andrej Karpathy, who is the publisher and first-person author. The Opus 5.5 and Claude roles are reported as his account, not as verified fact. > So this morning (Saturday) I took a hold of my Opus 5.5 and gave it the task summary — uncertain: The facts are largely faithful, but the reading flattens Karpathy's voice and an important condition. It reduces 'enshittified wall of slop' and 'idiocracy-grade content' to 'the same tired topics,' omits 'I can breathe again,' and drops his concession that the algorithm likely works for the median user but not for him. It also never says plainly that Opus 5.5 is a Claude model, so the shift from 'Opus 5.5' to 'Claude' may confuse lay readers. Smallest repair: add one clause with his median-person caveat and his stronger wording, and identify Opus 5.5 as Claude's model. > I'm sure this algorithm works well for the median person on the internet, but it definitely does not work for me. date — pass: originalPublished (2026-10-31T17:00Z) matches the post's own date line, 31 Oct, 2026, and the feed entry. The Friday-night and Saturday-morning sequence also matches the text. > HackerTube 31 Oct, 2026 support — uncertain: The stated claims are supported, but the quality interpretation and limitations understate what the post says about selection. The post gives at least an implicit basis: videos recommended by a few HN commenters, with Sonnet run over the comments in a Workflow to remove spam. The claim that there is 'no account' of how quality was evaluated, and the question built on it, should acknowledge that this crowd-recommendation criterion was stated even though it was not evaluated. > ran Sonnet over them in a Workflow (Claude Code feature), got rid of spam, and narrowed everything down to 40K comments recommending a YouTube video media — pass: The selected media is the verified direct link to Karpathy's original blog post, which is the most direct way for readers to reach the source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/35af159882f101d233556fcc1d6a9e1ef17615f229a0954857aa529db7c03d4a/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/hackertube-karpathy-builds-a-youtube-clone-from-hacker-news-35af1598.json)