Karpathy launches HackerTube, a Claude Code-built video recommender
cnBeta reports that Andrej Karpathy released HackerTube, a weekend project that uses recommendations found in two decades of Hacker News comments to assemble a video library. Karpathy says Claude Code, running Claude Opus 5.5 with medium reasoning effort, analyzed 43 million comments, selected 6,454 repeatedly recommended videos, and assisted with development and deployment. He also says the project took about two hours, while HackerTube requires no account and stores usage state in the local browser session.
HackerTube is a concrete demonstration of Claude Code handling large-scale data analysis, application development, deployment, security work, and performance optimization in a personal software project.
前特斯拉AI总监、OpenAI联合创始成员安德烈·卡帕西（Andrej Karpathy）近日发布了一个名为HackerTube的周末项目。他因为不满YouTube的算法推荐，利用Anthropic旗下AI编程工具Claude Code，从科技社区Hacker News过去20年的4300万条评论中筛选出6454个被用户反复推荐的视频，并搭建了一个全新的视频推荐网站。整个过程仅耗时约2小时。
卡帕西在社交平台上表示，相比YouTube现有的推荐机制，HackerTube能够提供“明显更好的观看体验”。
与依赖用户观看历史、个性化推荐算法的视频平台不同，HackerTube的核心思路是利用真实社区讨论记录，寻找那些经过长期推荐和认可的视频内容。
该项目的数据来源是Hacker News过去20年的评论。卡帕西让Claude分析其中4300万条评论，最终筛选出6454个持续获得推荐的视频，以此建立一个精选视频库。
#HackerTube#Claude Code#Claude Opus 5.5#Anthropic#Hacker News#YouTube#Google Cloud