Principal Component Analysis

Give a pgvector table a vector(3072) column for OpenAI's text-embedding-3-large, load a few million chunks, and the index from Chapter 7 refuses to build: CREATE INDEX ON chunks USING ivfflat (embedding vector_cosine_ops) WITH (lists = 1000) stops with column cannot have more than 2000 dimensions for ivfflat index. pgvector's documentation (version 0.8.6, as of September 2026) lists the ways out. Store the values as half-precision halfvec, which can be indexed up to 4,000 dimensions. Quantize them to bits. Index a leading slice of each vector, for models trained so that a slice still works. Or reduce the number of dimensions. For a model that was not trained to be sliced, the standard way to do the last is principal component analysis.