From the course: Tech Trends

Prompt engineering

From the course: Tech Trends

Start my 1-month free trial

Prompt engineering

“

- I'm Xavier Amatrian. I'm VP of Engineering at AI Product Strategy here at LinkedIn and I lead generative AI initiatives across product and engineering. In the context of generative AI a prompt is simply what you use to communicate with the AI and it can be as simple as a question and as complex as a very complicated construct that includes a lot of different components. So generally speaking a prompt can have any of the following, instructions, a question, input data, and examples. So what is prompt engineering? Prompt engineering is a new discipline that is appearing nowadays, and it's very hard to define because it's being created as we speak but it includes all the necessary components to manage prompts at a certain scale. And as I said before, a prompt can be very complex, right. It can be as simple as a question or it can be a set of examples. It can have instructions. It can have code. It can have all different kinds of templates. And if you need to manage that at scale because you're building a product that has an AI on the backend, that needs to be engineered. All that goes into engineering those prompts at scale, it's what we call prompt engineering. We need to also make sure that we understand the difference between the prompt creation aspect and then all the other engineering components that go with it. One of them is, for example, QAing. I mean, you need to QA prompts. You might have different variations of prompt that at scale you want to understand their quality and you're going to have some feedback loop that involves having maybe humans in the loop evaluating different aspects and different component of the prompt and then feeding that back into the engineering and the design of the prompt itself. Prompt engineering is a component of generative AI and I think it's going to evolve very quickly and very fast. And in fact, it's going to be leveraging itself which that is also something that it's mind boggling, but there's already a few experiments going on with AI prompting an AI, right. That's something you can do. You can prompt a language model so it generates a prompt that then goes into a text-to-image generative AI model to generate an image. So you can imagine that there's going to be AIs talking to themselves, and that's going to be part of the prompt engineering near future. But all of that, again, when needs to be managed at scale as part of a product, there's a lot of constraints like how do we protect, how do you make sure that people don't mess around when talking to the AI and making go off the rails? How do you QA and make sure that the quality is the right one? How do you productionize it and keep it working at a certain scale? I think that is all going to be generating a new discipline, really about the whole managing of those AIs and the way we talk to them. I think generative value is going to be revolutionary and transform our industry and probably our world in many ways.

Contents