HackerTube: YouTube built from videos recommended by HN
Andrej Karpathy has launched HackerTube, a novel video platform populated exclusively by videos linked within Hacker News comments since 2007. This curated alternative aims to combat the "enshittified wall of slop" often found on mainstream YouTube, offering a high-signal feed for discerning viewers. Its popularity stems from the creator's renown, the ingenious use of HN's collective intelligence, and the universal desire for better content discovery.
The Lowdown
HackerTube, the latest project from Andrej Karpathy, emerges as a minimalist video platform directly challenging the algorithmic malaise of mainstream YouTube. It meticulously curates its entire video library from links shared within Hacker News comments, spanning 17 years of community-vetted content.
- The platform sports a familiar YouTube-like user interface but presents a unique dataset of 6,454 videos, extracted from an analysis of 39,834 HN comments since 2007.
- Karpathy's motivation stemmed from a personal frustration with the degraded quality of recommendations on traditional video sites, seeking to create a higher-quality viewing experience.
- For those interested in deeper analysis or building their own projects, Karpathy has thoughtfully provided a downloadable .zip file containing the complete underlying video data.
HackerTube represents a refreshing, community-driven approach to video consumption, offering a curated window into the content deemed most valuable and interesting by the Hacker News community, free from opaque algorithms.
The Gossip
Algorithmic Ailments & Curation's Cure
Many commenters expressed deep dissatisfaction with YouTube's recommendation algorithm, often describing it as repetitive or low-quality. They contrasted this with the perceived superiority of platforms like TikTok's algorithm or the curated nature of HackerTube. The discussion highlighted a strong desire for more intentional content discovery, with HackerTube praised as a potential antidote to the 'enshittification' of mainstream video platforms. Some even mused about systems for sharing curated algorithmic profiles.
Project Particulars & Practicalities
Users were keen to understand the technical underpinnings and economic cost of building HackerTube, especially given its high-profile creator. Karpathy clarified that the project involved significant token usage for comment labeling (504M input, 7.6M output tokens, primarily cached Sonnet calls) but emphasized that the core application development itself was modest (~1M Opus tokens), making it a surprisingly inexpensive endeavor by modern LLM standards. Interest also arose regarding the availability of the raw data, which Karpathy confirmed was downloadable from the site.
Praise & Potential Projects
The overwhelming sentiment in the comments was one of appreciation for the clever concept and execution of HackerTube. Users expressed immediate enthusiasm for exploring the curated content, with some quickly identifying high-value videos like Alan Kay lectures. The project also served as an inspiration, prompting others to share similar personal curation efforts or discuss applying the same community-driven approach to other content domains, underscoring a broad appetite for alternatives to algorithm-heavy platforms.