vec2vec-performance-cosine-096-perfect-matching-8000

IN premise — summaries/2026/08/24/jha-2025-vec2vec-s1-introduction.md

Created 2026-08-24T17:10:57+00:00

vec2vec achieves cosine similarity up to 0.96 to ground-truth target-space vectors and perfect matching on 8,000+ shuffled embeddings without a pre-supplied candidate set.

Summary

vec2vec can correctly align over 8,000 shuffled vector embeddings to their true targets at 96% cosine similarity, all without being handed a list of possible answers in advance. This matters because it shows the system is genuinely solving the embedding-alignment problem on its own rather than doing a simple lookup, which means it can be deployed in settings where the target vocabulary is unknown ahead of time.