Linkup hosts retrieval debate: SPARSEUP, MICE cross-encoders and Cohere's v5 IR lessons
antoine_chaffin · x · 2026-11-02
Linkup Research hosted the first "Field Notes: On Retrieval" meetup at its Paris HQ, drawing 30+ researchers and engineers from Cohere and Sorbonne Université to argue about sparse vs dense retrieval for 6+ hours — quipping that "retrieval is solved" is something nobody in their office would say.
Three talks on the agenda:
- SPARSEUP — Linkup's head of research Thibault Formal introducing more competitive open sparse models;
- MICE — Sorbonne PhD student Victor Morand on minimal-interaction cross-encoders for efficient re-ranking;
- Cohere v5 IR Suite — Cohere's Carlos Lassance on lessons from shipping search models.
A second edition is coming.
More from Research
- SemiAnalysis launches GDPx benchmark — best model scores just 12% — generativist · 2026-11-02
- Two New Optimization Papers: Optimal SGD Stepsizes and Stochastic Rayleigh-Ritz Ascent — prof_grimmer · 2026-11-02
- Retinal photo foundation model RetiME predicts mortality risk across 4 cohorts — EricTopol · 2026-11-02
- Polytopia Bench: 12 Frontier Models Get 40 Minutes Each to Write a Rust Bot From Scratch — GregKamradt · 2026-11-02
- LiteFold goes generally available with Hill Climb Biology — sidahuj · 2026-11-02
- Pycea brings single-cell lineage tracing into the scverse ecosystem with TreeData — anshulkundaje · 2026-11-02