RAGs = the future of AI

Author

Andrei-Ovidiu Ersen
June 09, 2024

#3

Why you have to understand AI and RAGs?

Understanding AI will be essential for automating your business with AI.

ChatGPT and prompting can help a lot.

But having something personalized for your tasks is a game-changer.

RAG (retrieval augmented generation) AI agents are some examples of this personalized solutions.

RAGs is like giving the AI access to your documents or to websites.

So it is very useful to use this technology.

But you have understand what to use. Some RAG technology can be:

-too slow;

-not accurate;

We must find an equilibrium between the possible solutions and you must make the best decision for your specific case.

  1. Simplest case

The easiest way of creating a RAG is just to make an AI that can read through some documents.

This solution would be fast, but it may not be completely accurate.

It is nonetheless pretty good if you want just to find relevant information in those documents.

  1. When you should use something different

If the AI you use has to do a multistep process, then the simple RAG will fail. By multistep process, I mean calculating some trends (increases in sales in your business, for example) based on the information you provided in some documents.

Here we need to recreate the thought process of a human. In the image above, it is represented an AI agent that looks in some documents, selects only the relevant documents and then returns an answer. If no document is relevant, it performs a web search. Then it checks if the generated answer actually is relevant for the question. As you may know, AI can just make up stuff (these are called hallucinations). But with this method, of checking the answer, we can eliminate the hallucinations and drastically improve the accuracy of the AI agent. This comes of course at the expense of speed.

  1. Why this is relevant to you?

In the near future, there will be many people selling AI solutions to you, or many no-code tools that you can use to build your own solutions. By learning how a RAG should be designed, you will know exactly what you need, without going into the technical details of coding.

You must understand that to design a correct RAG you have to decompose the thought process that is happening in your brain and replicate it for AI. I will show you more examples in the future. But for now, you have to understand that AIs without this pre-programmed thought processes are still pretty dumb.

But you are pretty early and motivated to learn about AI (that’s why you subscribed to this newsletter) so congratulations.

Until next time, you can try my AI chatbots. These are the cheapest AI ChatBots on the market. They use your custom-made Databases. They have access to internet and to a database created by the user by just uploading .txt files and PDFs in a folder. It requires a Serper API (free account) and an Open AI API (2-3 $/month most likely). You can now skyrocket your learning, creativity and productivity with AI!

See you next week,

Andrei