#32 - Optimizing Interactions with GenAI
Using Everyday Communication Tools to Get the Best Out of GenAI
Intro
Usually, I dive into a technical concept and apply it to real-world scenarios to improve productivity and collaboration. This time, I’ll do the opposite.
Have you ever had an interaction with another person that wasn’t optimal? Maybe an argument went sideways, or a meeting became too emotional and loud? Or perhaps you just didn’t understand each other? Let’s see how possible resolutions for such cases can improve our interaction with GenAI.
We will cover:
Adding context.
Starting from scratch.
Using a different medium.
Using a mediator.
Let’s begin.
Possible Strategies
Adding Context
Do you know those people who come out of the blue and ask you something like, “Do you know why it behaves like that?” and you politely try to explain that you’re deeply out of context? Alternatively, you must decide on a topic you don’t have a complete perspective on.
The same applies to GenAI. The only difference is that, unless specified, it won’t ask any questions and will do its best to provide an answer. With minimal context, these answers are prone to mistakes and poor results.
Tips -
Always start with context, as much as you can.
Context can be text, images, and documents.
Specify to clarify and ask questions if needed.
Here is a brief example of metrics for a feature before and after sending the PRD -
Starting From Scratch
In some cases, a meeting or conversation just doesn’t get anywhere. You feel like you’re talking to a wall. Or maybe you’re in a three-hour-long meeting, and it’s just time to stop and start over at another time.
This might happen with GenAI as well. Long chats with significant contexts might get stuck or reach a point where optimal results are challenging. Starting over is an excellent possible choice.
Tips -
Clarify your goal and intent.
Ask to start over (e.g., “Forget everything so far and focus only on this prompt”).
Start a new chat (copy the original chat to keep some context in place).
Here is an example where ChatGPT struggles with generating a graph, but starting from scratch with several strategies eventually does the trick -
Use a Different Medium
A picture is worth a thousand words. In today’s business world, a meeting must include a slideshow or a few graphs. We usually use several mediums to ensure our message is clear before, during, and after a meeting or conversation.
The same applies to GenAI. Don’t stick to text only; if you feel stuck, add additional layers and mediums until you get the desired results.
Tips -
Start with a text prompt and evolve from there.
Look for quick wins with existing documents you already possess.
Screenshots and images can shorten the time needed to generate a prompt.
Here is an example where I use ChatGPT to generate a quick PRD intro using the prototype screenshot -
Use a Mediator
I have worked with stakeholders with whom I felt I always needed translators. I was never able to express myself well or fully understand their point of view. In those cases, it made sense to bring someone who could assist in moderating the conversation.
This is harder than the other options as it points to vulnerability with someone and requires trust in the mediator and the environment you can create in such a conversation. However, luckily for us, AI makes things easier.
The idea is simple:
In one chat, you work to reach your goals.
In another, you consult on the best prompt to send.
You can think of it as a manual implementation of autonomous agents.
Here is an example where, in one chat, I’m asking for help in analyzing a data set, while in the other, I’m consulting on the prompts I should send -
Summary
Think of GenAI as you think of other human interactions. While it isn’t human (just yet), practical tools and practices from our day-to-day interactions can be applied to improve the results we get from GenAI. By adding context, starting from scratch when needed, using different mediums, and seeking consultation, we can optimize our interactions with AI just as we do with people.