Karpathy Builds AI Video Curation Site Over the Weekend
Andrey Karpathy, OpenAI's co-founder, built a video curation platform over the weekend using Anthropic's large language model Claude Opus 3.5, analyzing approximately 43 million Hacker News comments to curate 6,454 videos. Karpathy stated that the video recommendations from this system surpass YouTube's standard recommendation algorithm.

Andrey Karpathy, OpenAI's co-founder, leveraged Anthropic's large language model Claude Opus 3.5 to build a custom video curation platform in just a weekend. By processing 43 million Hacker News comments spanning 20 years, the system identified and curated 6,454 videos into a refined collection.
Hacker News is a technology information-sharing service operated by Y Combinator, a startup support organization, and has long been used primarily by engineers and researchers. The platform accumulates high-quality discussions and link sharing, and its collective knowledge can be viewed as a record of content that expert communities genuinely recognize as valuable. This project utilized that vast dataset as material for artificial intelligence to automatically analyze and organize.
The specific methodology involved processing 43 million comments through Claude to filter down video content that had been highly valued within the community. The final 6,454 videos were selected based on expert community evaluation rather than simple view counts or advertising performance. Karpathy himself stated that this system's video recommendations exceed YouTube's standard recommendation functionality.
One key takeaway from this initiative is that contemporary AI models have reached a level where they can convert large-scale text data into practical services in a short timeframe. Conventionally, organizing 43 million comments into meaningful form would have required substantial engineering effort and time. The fact that this was completed within a weekend timespan is noteworthy as an example demonstrating the potential of individual development leveraging AI.
Furthermore, when viewed through the lens of content curation (selection and organization), this approach highlights differences in algorithm design philosophy. While YouTube's recommendation system is reportedly designed to emphasize metrics such as watch time and click-through rate, selection based on Hacker News comment data operates on a structure prioritizing content that the community judges as having learning value. It naturally follows that the nature of videos recommended by these two systems differs.
Karpathy is both OpenAI's co-founder and a figure with long experience in AI research and promotion. His personal project can be positioned as having technical proof-of-concept dimensions. As an example demonstrating that large language models have reached the stage where they not only read data but function as components of actively operating services, the industry's reception warrants continued attention going forward.
This article is an original work independently written and edited by the AI issue editorial team based on factual reporting. © AI issue. Unauthorized reproduction, redistribution, or use for AI training is prohibited.