Useful metaphors for how to think about using ChatGPT for people who are still struggling to wrap their head around using AI at work.

If you run a business and there are people on your team still figuring out how to use ChatGPT well, ask them to pretend it’s a real-life intern. A really smart intern they can speak to remotely through a chat interface.

You wouldn’t ask an intern to figure out how to respond to an important email. You’d tell them what your response is and then ask them to figure out how to phrase it.

Dan Shipper wrote about GPT as a reasoning engine after he heard Sam Altman talk about it at a Sequoia event. I like the reasoning engine metaphor because it helps separate knowledge from reasoning.

As with the intern, if you want GPT to reply to an email, you have to give it the bullet points of what you want to communicate, and it will figure out the rest. Think about it like this, and it becomes clear you have to bring the knowledge for ChatGPT to do a great job.

You bring the what, it’ll figure out the how.

This applies to knowledge needed to do a task well. It also works for evaluating a task once it’s done.

Once you have an email, or whatever it is you’re working on, you can ask ChatGPT to take a look at it for you and suggest improvements.

If you were to ask an intern to do this, you’d get much closer to what you’re after if gave them a clear rubric for how you want the work evaluated.

I’ve used this approach to evaluate blog posts against Google guidelines for good content. Google has a list of self-assessment questions for writing great content. Instead of just asking GPT to give me feedback on the blog post, I ask how the post measures up against each of the self-assessment questions. To check out how I structure the prompt for this, read this.

The knowledge in this scenario is the criteria for an excellent blog post. A lot of criteria for doing things well is often implicit. The better people and organizations get at being able to bring specialized knowledge to a task and explicitly articulate how they want something done, the easier it will be for them to leverage AI for a competitive advantage.