In-house multi-vec retriever crushes API dense retrievers on memory search, ColBERT next?
bo_wangbo · x · 2026-09-24
The author reports benchmarking several closed-source API-based dense retrievers against an in-house multi-vector model (not even tuned for memory search), finding a huge performance gap. He asks whether anyone has tried ColBERT / late-interaction retrieval for memory search, linking to the benchmark.
More from coding & agent
- Creator apologizes after AI agent auto-posted at 2am; revokes unsupervised posting permissions — LinusEkenstam · 2026-09-24
- Lovable launches Chats, detailing the agent architecture behind ~500M daily events — AlexandrePesant · 2026-09-24
- Xiaomi's MiMo V2.6 Pro builds a habit tracker in 62 seconds for under 5 cents — socialwithaayan · 2026-09-24
- Two engineers with AI shipped 3 finance infra projects in one quarter—after fixing what made agents guess — alex_verem · 2026-09-24
- How do you count an agent call blocked by an output guardrail? Real metering question — Rama_Surasani_ · 2026-09-24
- Brokerage MCP Postmortem: 4 Pitfalls Making Trader Leaderboards Safe for Agents — Accomplished_Fun_408 · 2026-09-24