Issue 2026-11-11 · Industry · 研究

Researcher Flags Benchmark Overfitting in GenDB-V Database Work

Thomas Neumann argues AI agents are driving overfitting in database benchmarking, calling GenDB-V's claimed 22x-61x speedups over Umbra and DuckDB infeasible. He attributes the gains mainly to per-query auxiliary structures precomputed at load time rather than to better general query processing.

Hacker News5 d ago
Read original ↗