LLMs learn dense retrieval from examples alone, no training required
Researchers have demonstrated that large language models can function as effective dense retrievers without any training, using only in-context examples to guide representation learning. RICE sidesteps the traditional requirement for supervised retrieval training by conditioning LLMs on contextual examples during inference, substantially outperforming prompt-based embedding baselines. This training-free approach expands the practical toolkit for practitioners building retrieval systems on commodity LLMs, reducing infrastructure overhead and enabling faster iteration on retrieval-augmented generation pipelines.
Modelwire context
ExplainerThe key omission from the summary is that RICE works by having the LLM itself generate embeddings at inference time, conditioned on examples, rather than relying on a separate retriever model. This is a shift in *where* the retrieval capability lives, not just a reduction in training overhead.
This is largely disconnected from recent activity in the space, which has focused on scaling retriever training and hybrid retrieval architectures. RICE belongs to a narrower conversation about whether foundation models can absorb retrieval tasks without specialization. The practical implication hinges on whether in-context conditioning is stable enough to replace fine-tuned retrievers in production RAG systems, which we haven't yet seen validated at scale across diverse domains.
If RICE maintains its performance advantage when tested on out-of-domain retrieval tasks (e.g., trained on Wikipedia, evaluated on biomedical or legal corpora), that signals genuine robustness. If performance degrades sharply on domain shift, the approach is primarily useful for closed-domain systems where examples and queries stay similar.
This analysis is generated by Modelwire’s editorial layer from our archive and the summary above. It is not a substitute for the original reporting. How we write it.
MentionsRICE · LLMs
Modelwire Editorial
This synthesis and analysis was prepared by the Modelwire editorial team. We use advanced language models to read, ground, and connect the day’s most significant AI developments, providing original strategic context that helps practitioners and leaders stay ahead of the frontier.
Modelwire summarizes, we don’t republish. arXiv cs.CL originally reported this story as “Effective Dense Retrieval using Only In-Context Examples”. The full content lives on arxiv.org. If you’re a publisher and want a different summarization policy for your work, see our takedown page.