RICE: training-free dense retrieval from LLMs using only in-context examples
_reachsumit · x · 2026-09-30
A University of Waterloo team (Nour Jedidi, Jimmy Lin et al.) introduces RICE (Representations from In-Context Examples), a training-free method that turns decoder-only LLMs into strong dense retrievers. By conditioning the LLM on a few query-document examples, RICE gives query and document encodings a shared context, substantially improving prompt-based LLM embedding accuracy. Code is open-sourced, offering a simple path to training-free LLM-based dense retrievers.
Related event: RICE Turns LLMs into Dense Retrievers Without Training(2 posts)→
More from Research
- HCOMP 2026 final session opens with FIBB paper on false information beliefs — windx0303 · 2026-10-01
- Meta-reasoning harness hits 71.5% on ProgramBench, beating Codex by 13.5 points — anirudhg9119 · 2026-10-01
- Rethinking inference-time scaling: what to compute, not just how much — anirudhg9119 · 2026-10-01
- Meta researchers propose Agentic Meta-Reasoning: letting models build their own reasoning graphs — anirudhg9119 · 2026-10-01
- FINGR: a dexterous hand solves a 2×2 Rubik's Cube with continuous finger tricks — HaozhiQ · 2026-10-01
- New lower bounds for kissing numbers: τ₁₉≥12268 and τ₂₁≥30761, open-sourced — felpix_ · 2026-10-01