WTF is RAG? It Could Solve LLMs’ Biggest Weaknesses

You may have noticed artificial intelligence researchers have a gift for coming up with new acronyms seemingly every week. We’re sorry to inform you of another one you need to remember: RAG, which stands for retrieval augmented generation.

It’s a method to help AI software such as chatbots answer questions about information they’ve never seen before. It could be the key to solving pesky issues such as chatbot software’s occasional habit of spitting out wildly wrong answers—dubbed hallucinations in the industry—or the rising computing costs to train the models that power the software.

Developers who want to do more with AI than buy a model from OpenAI or Anthropic have three choices: Build a model from scratch; tweak an existing model by giving it additional data so it can learn new information or a new task, a process known as fine-tuning; or retrieval augmented generation, which has surged in popularity.