BUILDING A WEB SEARCH ENGINE FROM SCRATCH IN TWO MONTHS WITH 3 BILLION NEURAL EMBEDDINGS

Summary from the TLDR newsletter

Search engines seem to be getting worse. Better quality results should rank higher than the prolific inorganic content and engagement bait everywhere today. Transformer-based text embedding models show amazing natural comprehension of language - this technology can be used to find great content, insights, and references. This article details a project where a developer tries to build a web search engine from scratch using neural embeddings. Embeddings appear far more powerful than traditional search.