Curation Makes a Comeback: HackerTube and the Revolt Against Algorithmic Feeds
How this was made Verified AI
Every Intellegix briefing is generated from that day's broadcast and run through automated checks before it publishes — with a human paged on any flag. Here is the trail for this edition.
The most-upvoted story on Hacker News on Monday, with 2,611 points, was Andrej Karpathy's HackerTube — a curated YouTube interface built entirely from videos that members of the Hacker News community have actually recommended to each other. Karpathy, formerly director of AI at Tesla and a longtime researcher at OpenAI, scraped years of HN comment threads to build a feed optimized not for watch time but for the revealed preferences of technically literate readers.
The project spawned imitators almost immediately. HackerBooks, posted by a user named roansh, scored 1,440 points and 241 comments applying the same logic to books mentioned on HN. Wikihack does it for Wikipedia articles. A project called HackerHack — essentially HN posts curated from HN, which the community found both circular and slightly funny — rounded out the ecosystem.
The HN thread framed the cluster as a rediscovery of curation as a product. Early internet directories like DMOZ and Yahoo's original category listings were replaced by algorithmic recommendation because it scaled — but scale came with costs now becoming apparent. HackerTube's implicit argument is that a smaller, better-curated corpus outperforms an infinite, poorly-curated one.
Karpathy described the project as a weekend experiment with no monetization plans, though commenters noted the underlying data layer — a structured record of what the HN community actually watches, reads, and references — carries real value for publishing, education, and content strategy. A light observer-effect concern surfaced late in the thread: the more widely known these tools become, the more they may distort the very behavior they measure.