Effective Large Language Model Adaptation for Improved Grounding and Citation Generation
Topics: AI Mode, AIOverviews, Chunk Relevance, LLM Readability, LLMO / GEO, Retrieval Augmented Generation (RAG)
This document introduces a framework called AGREE (Adaptation for GRounding EnhancEment) designed to make Large Language Models (LLMs) more factual and trustworthy. The method fine-tunes an LLM to generate answers that are grounded in provided documents and include specific citations for its claims. The framework also includes an iterative process where the model can actively search for more information to support parts of its answer that are not yet verified, improving the final response.
