Reranking Paradox: Performance Drops as Document Count Increases
CShorten30 · x · 2026-08-18
This post highlights a counterintuitive technical phenomenon in Retrieval-Augmented Generation (RAG): while the common workflow is "retrieve more documents -> rerank -> get better results," experiments show that Recall@10 actually plummets when pushing more than 100 documents to a Cross-Encoder reranker.
The reason is that rerankers act less like stronger retrievers and more like "boosting," amplifying the errors of the first-stage retriever. When input noise (irrelevant documents) becomes too high, the reranker is prone to misjudgment.
More from Infra
- Cursor releases Origin: a Git version optimized for AI agents — nicolascraske · 2026-08-18
- $10T in AI datacenter capex blocked: GPUs need 230GW, US grid delivers ~100GW — PeterDiamandis · 2026-08-18
- Power Grid Bottlenecks Stall $10T AI Data Center Buildout — PeterDiamandis · 2026-08-18
- Discussion: Running bots locally on Linux VMs — Daniel_Farinax · 2026-08-18
- PotatoMesh: Federated Dashboard for Visualizing LoRa Node Positions — tom_doerr · 2026-08-18
- llama.cpp adaptive MTP PR speeds up code generation by up to 100% — Look_0ver_There · 2026-08-18