Generation-Augmented Generation: A Plug-and-Play Framework for Private Knowledge Injection in Large Language Models
This addresses the challenge of efficiently integrating proprietary, fast-evolving knowledge into LLMs for specialized domains, offering a plug-and-play solution that avoids the drawbacks of fine-tuning and retrieval-augmented generation.
The paper tackles the problem of injecting private, domain-specific knowledge into large language models (LLMs) for high-stakes domains like biomedicine and materials, proposing Generation-Augmented Generation (GAG) to improve specialist performance by 15.34% and 14.86% over retrieval-augmented generation baselines on two benchmarks while maintaining general capabilities.
In domains such as biomedicine, materials, and finance, high-stakes deployment of large language models (LLMs) requires injecting private, domain-specific knowledge that is proprietary, fast-evolving, and under-represented in public pretraining. However, the two dominant paradigms for private knowledge injection each have pronounced drawbacks: fine-tuning is expensive to iterate, and continual updates risk catastrophic forgetting and general-capability regression; retrieval-augmented generation (RAG) keeps the base model intact but is brittle in specialized private corpora due to chunk-induced evidence fragmentation, retrieval drift, and long-context pressure that yields query-dependent prompt inflation. Inspired by how multimodal LLMs align heterogeneous modalities into a shared semantic space, we propose Generation-Augmented Generation (GAG), which treats private expertise as an additional expert modality and injects it via a compact, representation-level interface aligned to the frozen base model, avoiding prompt-time evidence serialization while enabling plug-and-play specialization and scalable multi-domain composition with reliable selective activation. Across two private scientific QA benchmarks (immunology adjuvant and catalytic materials) and mixed-domain evaluations, GAG improves specialist performance over strong RAG baselines by 15.34% and 14.86% on the two benchmarks, respectively, while maintaining performance on six open general benchmarks and enabling near-oracle selective activation for scalable multi-domain deployment.