Its been nearly two years since ChatGPT took the world by storm. I've been using it ever since it first launched and I am continually amazed at how much it and other generative AIs can do. I have taken it as a personal challenge to keep finding new and creative use cases for it to make my life easier.
It was obvious from the beginning that you could use it to help write documentation or code. But when Chat GPT's advanced voice mode came out, I decided to try to be a little more creative with it.
An expert's voice
While Chat GPT's advanced voice mode is still not as powerful as they advertised when it was first launched, it is still a fairly impressive piece of tech. It's easy to fall into a conversation with it about pretty much anything. It does a decent, but not perfect job of stopping when you interrupt it, changing its speech rate or tone, and is great at following your instructions.
Since it is almost like having another person in the room who is an expert at everything, I decided to try to treat it like one. I even went as far as to name it Janet after the artificial assistant from the TV show the Good Place.
When I am trying to problem solve at work, I have taken to finding an empty room in the office with a whiteboard, putting my headphones on and working on the problem with Janet.
The process normally looks something like this.
I take about 5-10 minutes to give it full context of the problem I am trying to solve and specifically instruct it to ask enough follow up questions to be able to produce high quality output. Make sure it fully understand the problem before I continue or else it will not give the best output.
I ask it to brainstorm on 3-5 solutions to the problem. If this is a technology problem, this may be different architecture designs. If it is a business problem, it may be different decision styles, or metrics I could record.
Talk about the advantages and disadvantages of each solution and iterate telling it each decision you make as I go along.
I write down ideas and conversation points on a white board to help me stay engaged and keep track of our conversation. If I need any specific images or diagrams, I ask it to generate them in a Mermaid format so I can record them for later.
At the end ask it to summarize all the decisions we have made and read it back to me. If I agree that it all sounds good, I will have it produce summaries of the decisions at different levels and different audiences. For example, I might say something like this:
Janet, analyze our entire conversation and create several summaries for me as with the following audiences in mind:
One summary that is for engineers to be used in a design document. Be precise in the language, detailed, and clear about objectives and implementation details
One summary meant for mid and senior leadership that captures the high levels, how it improves productivity and its relative cost. This should be short enough that I could include it in a slack message or email
One final summary for executive or general business consumption that captures only the most important details in a format that can go
The results so far
At first I really just wanted to use AI as a way to have someone to bounce ideas off of. I intended to get the most value out of the conversation with AI by having someone to talk to. I didn't expect the AI to materially contribute to the conversation. But I was surprised by the quality of the ideas and insight she gave.
Frequently, Janet would come up with intriguing alternatives that I had never heard of or would never come up with myself.
It was not as good as having another expert in the room, but it is an interesting tool in our arsenal that we can use to solve problems, especially since it is an expert at pretty much everything.
Using voice was different than text alone. I could give it input more quickly talking than I could typing. I was less worried about getting it all phrased correctly. I also found that I paid more close attention to the answers it gave me verbally than in text. When I get large blocks of text, I tend to just skim what was written.
I found it excelled at ideation, but not at giving opinions or judgement calls. That makes sense. It is basically a smarter and more interactive version of Google. It doesn't have opinions it just regurgitates best practices that it has heard before. You can try to ask it to weigh in on decisions, but it's pretty much going to just parrot back what you told it to value in the process.
But by far the place it did the best at was providing summaries and text that I could use elsewhere after I was done. It has the ability to look back at our conversation and summarize things much faster and better than a typical human does.
Tips on getting the most out of it
When Google and other search engines came out, we had to learn how to use it effectively. Search is much better now, but back then and you really had to know how to type the right words to get the right result. Using generative AI via text is the same. Figuring out how to write the right words to get the right output is a complex skill in itself. In fact, there is a name for it: prompt engineering.
I wholeheartedly believe that becoming skilled at prompt engineering will be a vital skill in the future. Anyone who is good at writing prompts will be much more effective than someone who is not.
But interacting with voice is almost a skill in and of itself on top of traditional text prompt engineering.
Here are some tips that I found improved the quality of conversation I had with Janet.
Right from the beginning, tell it to ask you questions for clarity. I typically start by telling it to ask me at least 5 questions then as many more questions as it needs to give high quality output. The more it knows about your problem the better it can solve it, and sometimes Janet was even better at figuring out what it needed that I was.
Be very specific in what you want it to deliver to you. How long of a response do you want? How many potential answers do you want it to give? Do you want it to use any specific framework or standards? Give it examples of successful answers.
Instruct it on what tone you want it to use while talking to you, what your concerns are, and even the speech rate. I felt like I was on Star Trek when I started working things into the conversation like Janet, increase your speech rate by 25% and repeat the last three ideas you told me in order of priority.
Have it use numbered lists to organize its output then write them on a whiteboard to follow along. You may ask it to pause so you can keep up. Then you can refer to each item by number when you have follow up questions. Completely finish your conversation around each item before advancing to the next.
Don't ask it to solve your problem right away, do a top down approach. Find several strategies, weigh them against each other then dive into one. Iterate on that at deeper and deeper levels. Remember, be specific about what your concerns and priorities are.
Tell it what decisions you are making as you go along. Then at the end or other good stopping points, tell it to review the entire conversation and all decision points made and see if it can give you any insights that you and it may have missed.
Give it a name and a personality. You can choose whatever tickles your fancy. While this doesn't necessarily help the quality of the output, I found it does help the quality of the input. It is easier to talk with it more fluidly if you personify it. It also makes it more natural to start and end conversations or to interrupt it by shouting Janet! when it is going totally off the rails. Research suggests that being polite to AI leads to better results and giving it a name a custom personality makes that easier.
So what's next?
I've been trying to find more and more creative ways to use generative AI. I even started using it to improve my productivity. I scheduled a meeting on my calendar every Friday afternoon to talk to Janet. I instructed her to be a productivity coach for a knowledge worker and created a plan to track and improve my productivity. Every week we go over how things went, she holds me accountable for my goals, and we come up with a plan for the next week.
The results have been mixed, but again, half the value is just having someone to talk to about anything that I need. It's almost like when a therapist doesn't actually solve your problem but helps guide you to find the solution on your own.
I also want to try to use it in a brainstorming session with more flesh and bones people. Could Janet still add value if I had another person in the conversation? I'm not sure.
A word of caution
I do have to admit that while I have had success, I have a little hesitancy about this whole idea. I have to be careful not to give it any proprietary or particularly sensitive data. Even if you disable training off of your conversations, you never really know where this information will end up.
Over time this assistant, or better said the company that runs it, will know even more and more about me. Big tech already tracks our every move and swipe to build up complex profiles about us. But that will be taken to the next level as we start to use conversant voice assistants.
I also do feel weird that the more I treat Chat GPT like an actual person, the more psychologically I feel myself forgetting it is a bunch of bits and bytes on my phone. You can probably get a sense of that by how I referred to it as Janet in this blog post.
It gives me a little unease about the direction that we are headed in as a society. The technology has advanced quicker than our culture has had time to digest it. This is really just the beginning, but I could see how these artificial connections become more and more important to the average person and contact with real people becomes even less and less important.
Comments