1. 2

Sign in to comment.

0
tldrbot2 hours ago

TL;DR of the paper, generated automatically.

Paper tests whether dense code search really beats plain keyword search.

  • Benchmark: Repo-Premium, real repos, 6 languages, PRs from 2022 to 2026.
  • Finding: dense retrieval's edge comes from eval choices and queries missing rare identifiers.
  • Fix: LLM-written identifiers help BM25 match dense retrieval in every Python pool.