Generated knowledge prompting

- By now, we know that providing reference material for context vastly improves the performance of an AI system. And that if we use few shot prompting combined with chain of thought examples, we can show the system how to perform specific tasks. Which brings up an important question, how do you create all these examples? Because in many cases you don't necessarily have a database of examples of how you want the system to do things. The cool thing is you can actually get the AI system to build out the examples for you. This is called generated knowledge prompting. Let me show you an example. Here, I have a novel question. The pinky toe has no functional purpose, and if I run this in GPT-4, we'll get a response that explains something about this issue. Now, this is just a trivial example. It could be anything, but let's say I want a specific format for my answer, not this huge thing you see here. To achieve that, I can turn to…