AI wars; Rag vs fine-tuning

AI wars; Rag vs fine-tuning

Strategy
Posted on
January 30, 2023
AI wars; Rag vs fine-tuning

AI wars; Rag vs fine-tuning

If you want

to personalize, institutionalize or specialize an LLM model. There are 2 most common ways to do this.

Rag (retrieval augmented generation) and fine-tuning.Let's explain the difference between these two with a simple allegory.

Think of a student,

we want to train this student as an engineer.Fine-tuning is like giving this student 3-4 years of engineering education.

It costs time and money, but in the end, an engineer is obtained. He answers all the engineering questions you will ask quickly with the information he learned.In other words, an LLM is obtained by training in engineering information and an engineer LLM.

As for Rag,

instead of sending the same student to university for 3-4 years, it is like giving him all the engineering documents.

When necessary, this student answers the questions asked to him by finding the most relevant and closest answer among these documents.

This significantly reduces the time and money cost. It is not as efficient as fine-tuning, but it is more than enough to meet many needs.

Therefore, it would be more accurate to determine the choice between the two according to the needs.In my next article, we will evaluate the subject from a technical perspective.

AI article series: 2

Previous article: How to incorporate AI into your company? Marketing company case study.