Using AI at work
What is retrieval-augmented generation (RAG)?
Also called
- RAG
- grounding
Definition
A method where an AI tool first looks up relevant information from a trusted source, such as your documents, and then uses it to answer.
Key takeaways
- RAG has an AI tool look up relevant information from a trusted source, such as your documents, before it answers.
- It lets a model use current or company-specific information without retraining, and makes it easier to show where an answer came from.
- Answers are only as good as the documents it searches, so outdated files lead to outdated answers.
In plain English
RAG helps a model answer from current or company-specific information without retraining it, and makes it easier to show where an answer came from. The answers are only as good as the documents it searches, so outdated files lead to outdated answers. Example: a customer service assistant that searches your latest price list and policies before replying to a customer.
A small business example
Picture an online store whose customer service assistant searches the latest price list and returns policy before replying. When prices change, updating the documents is enough, with no retraining needed. The weak point is upkeep: if last season's price list is still in the folder, the assistant may quote it.
Questions
How is RAG different from fine-tuning?
RAG looks up information at the moment of answering, so updating your documents updates the answers. Fine-tuning changes the model itself through extra training, which takes prepared data, technical effort and upkeep.