SignalTube is not affiliated with Andrej Karpathy or HackerTube.

PROJECT BACKGROUND · UPDATED NOVEMBER 3, 2026

What Is HackerTube? Karpathy's HN Video Site Explained

HackerTube is a project by Andrej Karpathy that makes videos linked in Hacker News discussions easier to discover. It turns thousands of scattered comment references into a searchable archive with topics, recommendation signals, and paths back to the original conversations. SignalTube is a separate catalog interface built from a public export, with its own routes, summaries, and browser-based library features.

This page explains what the project does, what its public build notes say about how it was assembled, and how to interpret its ranking signals. It also distinguishes dataset counts from the smaller set of videos SignalTube currently indexes. For a quick browse, start with all topics, popular Hacker News picks, or the dataset field guide.

What is HackerTube?

HackerTube organizes YouTube links that appear in Hacker News comments. The archive is useful because recommendations are otherwise difficult to recover: someone may mention a lecture in a programming thread, a documentary in a history discussion, or an explainer years after its upload. Topic labels create another route through the collection, while recommendation counts help surface videos that recur across different conversations.

The public export describes 6,454 video records, roughly 40,000 recommendation-comment records, 150 topics, and 16 broader areas. Its Hacker News history spans 2007–2026. Those are dataset measures, not platform-wide totals: the export reflects the captured archive and its classification rules. SignalTube's current public video list follows 6,386 sitemap entries from the source snapshot. Differences come from the indexed list and filtering; they should not be read as deleted Hacker News records.

How Karpathy built it

Karpathy's project write-up describes a compact, roughly two-hour build. It names Opus 5.5 and Sonnet as coding agents used during the work, with the broader workflow examining around 40,000 Hacker News comments and producing a catalog of 6,454 videos. The resulting interface includes 150 topics grouped into 16 areas. These figures come from the project description and the public export; they are not estimates generated by SignalTube.

The high-level workflow is straightforward: collect candidate YouTube links from Hacker News, group references by video, label each mention by its role in context, then organize the resulting video records with topic metadata. A source export makes that work reusable: downstream sites can create alternate search interfaces without rebuilding the entire historical crawl. SignalTube uses the export as input and keeps a separate database, routes, and page copy.

How the ranking works

A recommendation archive answers a different question from YouTube's trending page. It uses Hacker News discussion activity as a discovery signal. Distinct recommenders and repeated mentions can indicate that a video was useful to several readers or returned to conversation over time. Topic membership narrows the pool; a time range shows when the link appeared in Hacker News.

Counts need careful reading. A mention is not automatically an endorsement: the source distinguishes strong recommendations, ordinary recommendations, contextual references, critical mentions, and jokes. The count of distinct recommenders is also not a view count. A video's first and last Hacker News mention years do not identify when it was uploaded. On SignalTube, the “best” list follows the source recommendation score, while topic pages use a separate recency order; the page labels make that distinction visible.

HackerTube vs SignalTube

Both sites help readers find videos mentioned in Hacker News, but SignalTube is an independent implementation. It reuses the attributed public data export while presenting a separate browsing experience. The current catalog suppresses source summaries and long comment text, writes its own metadata-based video guides, and links readers to each Hacker News comment for the original context.

How to use the dataset

Start by choosing the task: discovery, analysis, or building a new viewer. For discovery, open a topic archive or a best talks list. For analysis, inspect the video, category, mention, and related-video fields and keep each measure distinct. For a viewer, preserve stable video IDs, source comment URLs, and the required attribution. If an item has no verified duration or upload date, leave that property empty instead of inferring it from its Hacker News history.

Reuse also requires attention to rights and freshness. The catalog data is attributed under ODC-By 1.0, while individual comments and videos retain their own authors and platform terms. Current YouTube availability can change independently of the archive. SignalTube's build notes guide, alternatives overview, and top talks list provide additional ways to explore the index.

Frequently asked questions

Continue with the dataset guide, browse the topic index, or open a featured video page.