Researchers have proposed ICICLE, a new framework for in-context document retrieval that aims to improve the efficiency of adding new documents to a database. The current method of generative retrieval requires updating model parameters and retraining when new documents are added, which can be costly. ICICLE uses context-provided information to generate document identifiers, allowing it to expand the corpus without retraining. Experiments show that ICICLE improves retrieval for new documents while retaining performance on existing ones. --- Why it matters: This matters because current generative retrieval methods have limitations when dealing with large-scale datasets and frequent updates. ICICLE's approach could help alleviate these issues by enabling more efficient addition of new documents, which is crucial for applications like search engines and question-answering systems. Source: https://arxiv.org/abs/2605.26902