Andrej Karpathy built a curated video platform in a weekend using Claude and 20 years of Hacker News

What changed

Andrej Karpathy, co-founder of OpenAI, built a curated video platform over a single weekend. He used Claude’s Opus 5.5 model to sift through 43 million Hacker News comments spanning 20 years. This massive data set was filtered down into 6,454 carefully curated videos that Karpathy claims outperform YouTube’s native recommendation algorithm.

Why builders should care

This approach shows how modern AI models can accelerate content curation at an unprecedented scale and precision. Instead of relying on broad watch time or click metrics, leveraging a rich, domain-specific historical text archive combined with a capable AI can produce highly relevant video suggestions. Builders working on recommendation systems, content platforms, or knowledge discovery can adopt similar data-driven filtering to increase engagement and content quality, cutting through the noise that typically plagues large video platforms.

The practical takeaway

Karpathy’s method stresses the value of combining deep domain data with powerful AI as a shortcut to better recommendations without needing massive proprietary metadata or complex infrastructure. For founders or operators, this means curated AI-driven video or content products can be launched quickly and scaled using existing public data sets and off-the-shelf LLMs. Much of the competitive edge shifts from pure engineering scale to smarter data preprocessing and model prompting.

What to watch next

The key development is how foundational models like Claude 5.5 reshape content moderation and recommendation layers by efficiently digesting conversational and archival text alongside video metadata. Watch for new consumer platforms or enterprise tools that combine large public or community datasets with AI to outsmart algorithmic gatekeepers like YouTube. Also, the question will be whether AI curation can maintain user trust and transparency where opaque algorithms have fallen short.

AI Quick Briefs Editorial Desk