She tested it again. “Show me ‘crew morale reports, last quarter.’” The vectors spread out: happy performance reviews, tense shift logs, a birthday party recording. Then, a gap. A cold, lonely gap. Beyond it lay the maintenance logs for the waste recyclers—technically adjacent in time, but not in meaning . Autocut sliced cleanly. The waste logs were excluded.
Elara leaned back. The cluster hummed, satisfied. weaviate autocut
: If you set limit: 5 , you get five results, even though the last two are significantly less relevant than the first three. She tested it again
autocut: 1 : Returns only the first group of results before the first major jump in distance. A cold, lonely gap
: By filtering out irrelevant results before they reach expensive re-ranking models or cross-encoders, you save on computational costs.