Rag

This post is primarily based on the survey “Retrieval-Augmented Generation for AI-Generated Content: A Survey”. I presents retrieval-augmented generation (RAG) in six parts: background, method, enhancement, applications, outlook, and takeaways. Background In recent years, we’ve seen a rapid surge in Artificial Intelligence Generated Content (AIGC), driven by large generative models that can produce text, code, images and even videos (Zhao et al. 2024). For text and code, widely used examples including GPT-style models and Anthropic’s Claude family (Achiam et al. 2023, Anthropic 2024). For images, modern systems are often powered by diffusion-based text-image generation, including latent diffusion models (Ramesh et al. 2021, Rombach et al. 2022). For video, OpenAI’s Sora is a prominent example of large-scale text-to-video generation (OpenAI 2024). ...