We revisit retrieval-augmented generation (RAG) by embedding retrieval control directly into generation.Instead of treating retrieval as an external intervention, we express retrieval decisions within token-level decoding, enabling end-to-end coordination without additional controllers or classifiers.Under the paradigm of Retrieval as Generation, we propose GRIP (Generation-guided Retrieval with Information Planning), a unified framework in which the model regulates retrieval behavior through control-token emission.Central to GRIP is Self-Triggered Information Planning, which allows the model to decide when to retrieve, how to reformulate queries, and when to terminate, all within a single autoregressive trajectory.This design tightly couples retrieval and reasoning and supports dynamic multi-step inference with on-the-fly evidence integration.To supervise these behaviors, we construct a structured training set covering answerable, partially answerable, and multi-hop queries, each aligned with specific token patterns.Experiments on five QA benchmarks show that GRIP surpasses strong RAG baselines and is competitive with GPT-4o while using substantially fewer parameters.

In software development, testing is critical in guaranteeing software quality, with test case design at the core of the testing phase. However, generating effective test cases requires deep expertise and significant time and effort. Therefore, much prior research has turned to methods of Natural Language Processing, utilizing generative deep learning methods to automate test case generation. These earlier studies, however, have largely ignored the crucial role of conditional statements within software requirements - a factor we believe is indispensable for generating high-quality test cases. To bridge this gap, we introduce a pioneering approach for automatically deriving antecedents and consequents from requirements, termed as Traceable Sequence Generation (TSG). The TSG generates conditional statements first and then generates corresponding test cases by constructing a Cause-Effect-Graph. To verify the effectiveness of TSG, we constructed a requirement-to-test-case dataset, called Code Test Case Eval (CTCE). The dataset also includes annotated conditional statements for each segment of the requirement text, so we can utilize them to improve test case generation easily. Our experimental results indicate that TSG notably surpasses traditional and NLP-based methods, excelling in conditional statement extraction and generating high-coverage test cases.