I still remember the early days of YouTube, over 20 years ago. There was a dedicated tab for recommended videos, based on view counts and user comments. Videos across various genres and topics were listed—all driven by human interaction, without any algorithmic interference. I must admit, I miss that era; the recommendations were often much better aligned with personal tastes.
This is something that resonated with Andrej Karpathy, one of the co-founders of OpenAI. With the help of Claude, he developed "HackerTube," a platform that showcases videos deemed the most relevant on YouTube. Karpathy processed millions of user comments from the HackerNews website—spanning from 2007 to 2026—to identify the videos that were recommended most frequently.
Claude then compiled a list of 6,544 videos across 150 topics, ranked by how often they were suggested. These topics include science, software development, video game development, history, programming, engineering, and much more.
Beyond curating the most relevant videos based on HackerNews discussions, HackerTube also generates its own video thumbnails—derived directly from the video content itself, rather than using those uploaded by the content creators. According to Karpathy, this approach harks back to the early days of YouTube and avoids the AI-generated thumbnails that currently saturate the platform.
HackerTube does not require a login; viewing history is stored locally within your web browser. You can access it via the link below.
