aiqbench analysis
This research presents RICE, a training-free method that leverages in-context learning to generate high-quality dense representations directly from decoder-only LLMs. By conditioning the model on shared examples for queries and documents, this approach significantly boosts retrieval accuracy without requiring additional retriever training. It offers a practical pathway to enhance search capabilities using standard language models, reducing the computational overhead typically associated with specialized retrieval systems.
Effective Dense Retrieval using Only In-Context Examples
arXiv:2609.38099v1 Announce Type: cross Abstract: Turning decoder-only large language models (LLMs) into strong dense retrievers typically requires some form of retriever training. In this paper, we ask whether LLMs can instead be prompted to produce effective representations for dense retrieval given only a few in-context examples. To answer this, we introduce RICE (Representations from In-Context Examples), a simple "training-free" approach that extracts high-quality dense representations from LLMs. To do so, RICE conditions the LLM on examples that provide a shared context for query and document encoding. Our results demonstrate that RICE embeddings can substantially improve the accuracy of prompt-based LLM embeddings, establishing it as a simple method to build LLM-based dense retrievers that do not require training. We release our code at https://github.com/nourj98/RICE.
Abstract:Turning decoder-only large language models (LLMs) into strong dense retrievers typically requires some form of retriever training. In this paper, we ask whether LLMs can instead be prompted to produce effective representations for dense retrieval given only a few in-context examples. To answer this, we introduce RICE (Representations from In-Context Examples), a simple "training-free" approach that extracts high-quality dense representations from LLMs. To do so, RICE conditions the LLM on examples that provide a shared context for query and document encoding. Our results demonstrate that RICE embeddings can substantially improve the accuracy of prompt-based LLM embeddings, establishing it as a simple method to build LLM-based dense retrievers that do not require training. We release our code at this https URL.
Information Retrieval (cs.IR); Computation and Language (cs.CL)
arXiv:2609.38099 [cs.IR]
(or arXiv:2609.38099v1 [cs.IR] for this version)
https://doi.org/10.48550/arXiv.2609.38099 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 218 words
Related coverage
Introducing Gemma 4 models on Amazon Bedrock
AWS · Jun 15, 2026
Evaluating In-Context Learning and Retrieval Strategies for Devanagari Post-OCR Correction
arXiv · Sep 21, 2026
Article: Your Next DSL Author is a Language Model
InfoQ · Sep 16, 2026
Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers
Hugging Face · Aug 26, 2026