Errpedia benchmark: all 20 tested exact k-NN vector search setups return wrong top-10 results
_reachsumit · x · 2026-11-18
A new arXiv paper introduces Errpedia, a dataset for testing the numerical correctness of exact k-NN vector search, with ground truth independently verified via exact rational arithmetic.
- It contains 1,616 queries across 21 error categories; all 20 tested library and database configurations return mathematically incorrect top-10 results on at least one query.
- Failures include misranking nearly equidistant vectors and incomplete results: one library omits all 10 exact matches for a query, and several systems return fewer than k neighbors.
- The cause is floating-point rounding and overflow; the dataset is publicly released for developers to test their implementations.
More from Research
- Voyager 1 Reaches One Light-Day From Earth, Signals Now Take 24 Hours — snowclipsed · 2026-11-18
- TCS Hiring Statement Urges Quality Over Quantity as AI Fuels Paper Arms Race — fortnow · 2026-11-18
- New paper: a sociocultural approach to linguistic creativity — abenitezburraco · 2026-11-18
- 27B Model Rewrites Its Own Weights Without Training or Gradients, Paper Claims — udmrzn · 2026-11-18
- From 'impossible' to a working Bluetooth beacon: open-source EDA chip design is real — IgorCarron · 2026-11-18
- AURA 2.0: a streaming video LLM that speaks first, calls tools, and grounds visually — Shuhan Li · 2026-11-18