Prompt chaining is the way of using the output of one prompt from a generative AI model as the input to another prompt. It can be a powerful way to complete tasks that require multiple steps. But this idea of chaining doesn’t need to be limited to prompts. Integrating generative AI outputs into complex tasks that use many different tools and techniques can empower teams to work smarter and more efficiently than ever before.
Here’s how we took the 100 most engaging articles on generative AI captured from a media monitoring tool and used ChatGPT to identify key coverage topics and keywords that could be fed back into that monitoring tool for deeper insights.
First, we need a way to convert article webpages into a text file (more on that later). I’m not a programmer, but I do know that every Mac comes with a program called AppleScript to automate tasks like converting 100 webpages into a text file, and that ChatGPT is pretty good at writing code:
Do I understand how this script works? Sort of. Is this the best AppleScript to accomplish this task? I have no idea. But it works, and that’s good enough.
Second, we evaluate those articles using Perplexity – an AI service powered by ChatGPT but with additional functionality that OpenAI doesn’t offer. One of the most useful is its ability evaluate user uploaded information (hence the first step):
Let’s zoom in on the Generative AI and Art topic at #4:
Now let’s see the keywords related to the second subtopic on tools artists can use to protect their work:
Here we have a handful of highly relevant keywords: Specific tool names like Nightshade and Glaze and more conceptual terms like Data Poisoning and Pixel Manipulation.
Third, create a query with these new terms and feed it back into the media monitoring tool to find more relevant articles.
Repeat these steps to get even more keywords:
Chaining generative AI outputs with more traditional tools can empower teams to use those tools in new and more effective ways. In this example, a team could setup media monitoring with keywords derived from hundreds of articles in less time than it would take to read just a few of them. But this approach isn’t limited to media monitoring. Any tool or complex process that involves summarization, extraction, or evaluation of text can be enhanced with generative AI. All you need to do is start working outside the prompt.
– Eben Gilfenbaum