🧠Knowledge Series #42: What is RAG?

Retrieval-Augmented Generation explained. Why every product team needs to know more about RAG.

🔒The Knowledge Series is a collection of easy to read guides for product teams and tech / business professionals, designed to help you fill in the gaps in your tech knowledge so that you feel more confident when chatting to more technical colleagues. 

Clearly explained in plain English. One topic at a time. 

If you’re a free subscriber and you’d like to unlock them you can do so below. Or you can learn more about what you get with paid access here.

---

Hi product people 👋,

Unless you’ve been directly involved in developing AI-powered features, you may not have heard of Retrieval Augmented Generation (RAG).

But with commentators suggesting that RAG is now an essential part of the modern product tech stack and Amazon’s AWS and Google scrambling to release powerful RAG capabilities to help product teams improve the accuracy of their AI-powered features, RAG is quickly gaining momentum and is definitely one emerging technical concept to keep an eye on in 2024.

This Knowledge Series will explain exactly what RAG is, the role it plays in helping increase the reliability of AI-generated content and how product teams at companies like Twilio and Atlassian are using it. 

We’ll also look at some end to end examples and explore the essential terminology worth knowing so that if your company decides to implement AI powered solutions like chatbots, you’re fully equipped to understand the role that RAG can play in augmenting AI features.

Coming up:

• What is retrieval augmented generation (RAG)?

• Why is it important for product teams to understand?

• How are product teams using it in 2024? A look at some real world examples from Twilio, Atlassian and others

• How is it different from other AI-related technologies?

• RAG in practice: an end to end example

• Key terminology worth knowing 

---

The Knowledge Series

What is retrieval augmented generation (RAG)?

One of the most significant downsides of LLMs and generative AI is their tendency to hallucinate; that is, provide an incorrect answer to a question with such confidence that the user who asked it in the first place thinks it’s correct.

It’s one of the reasons why Google is understood to be reducing the frequency with which it is displaying its Search Generative Experience and is a major barrier to mass adoption of tools like Perplexity. If you have to double check the “answer” you confidently get from a search engine then why wouldn’t you just search the web and find an authoritative source instead?

This is where RAG can help.

Retrieval Augmented Generation is a fancy way of describing the process of optimizing the output of a large language model by asking the model to reference authoritative knowledge bases outside of its training data before generating a response.

To use a cooking analogy (as we often do in the Knowledge Series!), it’s a bit like asking a chef in a French restaurant to create a dish known only to a specific region in India and hoping they come up with the recipe you’re looking for. Without giving the chef a series of additional niche recipes and access to the books published in that specific region, the chef will struggle to create it. Instead, they’ll refer to their own library of recipe books and produce something which may look authentic but when you taste it, it’s not quite what you asked for.