Secrets of successful adoption

As we discussed in Chapter 1, LLMs can adapt to different domains. This means you can adapt an LLM to your specific task by providing instruction in the form of natural language. Therefore, sometimes, people tend to think that generative AI is different from “classical ML” and that you don’t need to meet any specific data requirements to start building your application. We don’t believe this is entirely true, especially not in an enterprise context.

However, you do need a high-quality dataset to build a successful generative AI application, although it could be smaller than the one required by “classical” ML approaches. And that’s probably the most common misconception that leads to unsuccessful POCs in this area. So, what kind of data do we need?

First and foremost, we strongly recommend that you begin by drafting a few examples of the desired input and output for the generative AI solution you&...