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.
Read original ↗