I Used AI to SpeedRun Wikpedia

A developer built an AI-powered Wikipedia speedrun solver using embedding models and cosine similarity. The tool encodes the target word into a high-dimensional vector, computes embeddings for all links on the current Wikipedia page, and greedily follows the link most similar to the target. It solved Ronaldo-to-goat in under 5 seconds and 14 links, compared to the author's manual 3 minutes and 11 links.