BREAK IT! (... down)

Make complex ideas, simpler. Make complex requests, simpler. In simpler terms, break it… down.

That is: break complexity down if you want better responses and results from using Generative AI (Gen AI) tools.

This doesn’t mean that Gen AI can’t process large amounts of data and text (tokens) as input and outputs, it just means that the quality of the response it provides may (not necessarily, however) decrease the more complex the request is.

I asked ChatGPT and Claude: “does the quality of your responses diminish the more complex the request is?”

The bottom line is that, while some level of quality may be preserved the more complex and nuanced (and yes, I am using the word “nuanced” that has now become a forbidden word in my writing since AI seems to be overusing it!) and large your request is, in my experience quality does decrease, the answers tend to become wordier, but less significant and relevant.

In addition, and I am not sure that this happens across all models and tools because I have not seen it, but it definitely happens to ChatGPT: there always seems to be a maximum length for an output (even though I asked ChatGPT about this and it said no, there is no maximum length), which means that if your request is too complex, the response will be as long as if the request had be simpler, and you may be missing some good content from AI.

Let me give you an example.

Suppose you want Gen AI to help you write a corporate policy or guidelines about how to conduct job interviews that are unbiased, fair, equitable, respectful and inclusive.

You can literally ask your preferred AI model: “write a corporate policy or guidelines about how to conduct job interviews that are unbiased, fair, equitable, respectful and inclusive” … I did it… you and go and try it, and see what you get as a result. I asked Claude and ChatGPT. Both of their responses were… lousy… to say the least.

Now, let’s try a different approach. Instead of asking your AI model: “write a corporate policy or guidelines about how to conduct job interviews that are unbiased, fair, equitable, respectful and inclusive,” let’s break it… down, and ask the model (in a separate chat, don’t use the same chat you used before for the original request):

“What are the most important sections that are a policy or guidelines about how to conduct job interviews that are unbiased, fair, equitable, respectful and inclusive, must include.”

And now you will get a different response.

And your next step can be to create a framework to start developing each section, one by one, starting with making sure that the first one is solid and you use it as a “sample” for the following sections of the policy.

For example, you can ask your Gen AI tool: “Act as a recruitment professional with expertise in diversity, equity and inclusion, and an advanced professional program in the neuroscience of biases. Develop the content of the first section in your suggested list. Make sure that the section addresses all the key components you outlined, it is no longer than 500 words, and it is written in the voice and style that I will share with you. Please confirm that you understand my request before proceeding and then I will share a sample of our policy writing style.”

Once you do that, you can share the sample and give the tool permission to either ask you questions to clarify, or to go ahead and write the first section of the policy.

You will continue to break down the complexity, and refine the writing of that first section. Once that section looks beautiful, the way you want it, you can now tell your AI tool: “Use this final write up of the first section as the sample to now develop the content of the second section.”

Try this, for all the sections… and compare the result to the original, complex, unbroken question. You will see the difference.

Try it with any other example of your interest, and let me know in the comments how it goes!

By Enrique Rubio (he/him)

(Note: I write fast, speak fast and read fast. Sometimes - often times - I don't see the typos! Please let me know if there are typos or grammatical misconstructions in my write ups. It happened to Shakespeare, it certainly can happen to me. Do so with kindness, compassion and grace, though. I appreciate it!)

Download Our Full Guide

Click here to download.