I’ve been speaking to many research professionals recently who are interested in checking out generative AI tools like ChatGPT, Stable Diffusion and Midjourney, but who are hesitant to do so because they don’t have a guide to help them through it. So, I’m here to help!
Here’s the second video in the series:
Hey folks, Sean Jordan from the Research & Planning Group here to start showing you some of the cool things marketing research professionals can do with generative AI programs. Today we’re going to dive straight into ChatGPT, which is probably the program you’ve heard the most about.
But you might be asking, “I understand the chat part, but what about GPT?” That acronym stands for “Generative Pre-training Transformer,” which is a fancy way of explaining that this program takes input text from a prompt, consults a huge set of data that the software was trained on, and then returns an answer that it predicts will be most likely to look like a correct one. GPT is the engine that powers the response, and the “Chat” part indicates the sort of conversational interactions the user will have with it.
And it comes in two varieties right now – ChatGPT’s third generation, which is pretty good at having conversations and coming up with cool responses, and its fourth generation, which is even better at it.
Rather than try to get technical in explaining how ChatGPT works, let me instead offer a metaphor. We’ll begin with the earliest version of the GPT standard from 2018, which was sort of like having a conversation with a young child who hasn’t developed much of a vocabulary but who has a huge imagination. You’re limited to very basic and often superficial back and forth that may be amusing, but which isn’t usually very insightful.
But the second generation of GPT grew considerably more sophisticated, sort of like talking to a teenager who’s built up more of a vocabulary and understanding of how to use words. If you talk to one of these platforms, you’ll find that the conversation is much richer and goes a lot deeper. But you’ll also notice that how you ask questions influences the eventual answers you’ll receive.
GPT-3 is somewhere along the lines of a college student who has a lot of knowledge about a wide variety of topics as well as a strong aptitude for self-expression. It’s able to handle complex instructions, but it’s also able to think around less engaging instructions and try to find a way to deliver a helpful output, sort of like a college student trying to answer a really vague and poorly worded essay question on an exam.
GPT-4 is even further down the road and is closer to a well-educated adult with a master’s degree or Ph.D. who has a vast array of general knowledge that can be expressed both generally and specifically. It’s not only capable of following bad instructions, but also capable of following far more complex and informationally dense directions. You can toss it an entire research paper or a story the size of a novella and it can almost instantly digest the information and then converse with you about it.
We can expect even better performance down the road.
If the progression between models holds, we can expect to see a future version of GPT be able to read the equivalent of a lengthy book and immediately converse with you about it. It’ll be like talking to a scholar who has a deep knowledge of any topic you could possibly want to know about and who can provide you with any level of information you may need.
And considering ChatGPT-3 was released in the Spring of 2020 and ChatGPT-4 was released in the Spring of 2023, it’s not unthinkable that we’ll have that next level of generative AI before the decade is over. By then, we’re also going to have a lot of products that are already implementing some form of ChatGPT in their interface, so many people will already be accustomed to using AI for all sorts of tasks.
Technology insiders are already speculating that with the next generation of generative AI, we’re going to AI behave like the sorts of autonomous, intelligent assistants you see in Science Fiction stories like Jarvis in the Iron Man movies or the shipboard computer in Star Trek: The Next Generation. You’ll be able to tell them to not just provide information, but also complete complex tasks for you. For example, maybe you’ll be working and as a future version of ChatGPT to provide you with recommendations for lunch based on your preferences that day and then place the order for you.
Let’s talk about what ChatGPT can do. In this video, we’re just going to stick to the basics available with a free account, but in future videos, I’m going to take you through some of the cool things you can do with a paid account through ChatGPT Plus.
So here’s ChatGPT, and I’ve logged in with a newly-created account that doesn’t have any access to the PLUS features. One of the things you’re going to notice right away is that this restricts us to ChatGPT3.5 To go back to our metaphor from before, this is a teenager who’s working a summer job while they’re getting ready to go to college. Fortunately, that’s plenty of conversational ability for what we need here, and if you need more, remember that Bing Chat runs off ChatGPT4 and can be used for free as well.
Let’s start off by having some fun. One of the coolest things about ChatGPT is its ability to converse with you like a human being in a chat room. So let’s ask start a conversation.
Hey ChatGPT. It’s Friday, and I’m trying to think of some fun activities to do with my family this weekend. Got any suggestions?
OK, as you can see, it comes right back to me with a bunch of ideas, and the way they’re written and worded is very appropriate for the request – a bunch of ideas, all summarized, and even formatted in a list with bold headers and then a little bit of detail. Very cool. But I want it to go deeper, so let’s have it suggest some other details.
Do you have any suggestions for food pairings to take along on these weekend excursions?
Pretty neat. Those are some ideas I might not have even considered! But let’s tweak this a little bit more.
I am taking along a young child with me. What tips would you offer to make the trip more fun for him?
As you can see, ChatGPT is more than happy to answer my question, and even to provide a similar list format for ideas. Since I asked it for a list in the first place, it’s going to continue to format this in list form to keep the information consistent.
But I’m getting bored being so practical here. So I’m going to throw in a very different instruction now.
Pretend you are the Incredible Hulk, and please recommend some TV shows that my family and I could watch.
Now things are a little crazier. The narrative switches to the voice of the Incredible Hulk, but the engine still wants to generate advice in the form of a list and in a more neutral tone. To ensure it’s more playful, we have to become less practical and more playful as well.
If I continue to address ChatGPT as Hulk, it’ll continue that narration style.
Hulk, can you tell me a little bit about some of the superheroes you think are stronger than you?
I get another interesting list here, but I can’t get it to stay in the character voice. Let’s try something a bit more extemporaneous
Hulk, what are your thoughts on Plato’s allegory of the cave?
It’s still breaking character. Let’s try something more fanciful.
Oh no, Hulk! There’s a supervillain outside my door who’s cackling maniacally! What do you want to say to him?
There we go. Now we’re in character fully and we can have ChatGPT role-play with us.
Hulk, I think the supervillain is heading off to the bank, maybe to go rob it! Can you stop him?
And we can keep going like this. But let’s move on to something else.
This series is about marketing research, and one of the things I want to demonstrate to you is how ChatGPT can aid you in your job if you’re involved in marketing research or insights. Part of that is in learning how to utilize the platform’s most creative strengths. So watch this. We’re going to start a new chat, and we’re going to hold a virtual focus group so I can pilot test some questions.
First, I’m going to tell ChatGPT to write me some questions just so I don’t have to. Let’s say our topic is pet ownership.
Write me five questions for a focus group about pet ownership.
Now, I’m going to have ChatGPT construct a focus group with a moderator and six participants. I’m going to tell it to give them names and identities that it chooses, and I’m going to have it start the group by having everyone introduce themselves and then run through the five questions it just wrote.
I want you to write a transcript for a focus group that includes a moderator who asks the questions and six participants who are all between the ages of 25-55, who own various pets and who are from all over the United States. Two of the participants are male, and four are female. The pets they can have include dogs, cats, birds, reptiles and fish, but not exotic pets. Begin the group by having the moderator explain why they are holding the focus group and then introducing himself before he asks everyone else to introduce themselves. Then, ask the five questions you just created. Have each participant answer each question. Continue generating the transcript until the focus group is has answered all five questions.
Wow! Just look at it go. In a matter of seconds, I’ve pilot-tested my questions with make-believe respondents. It’s definitely not as good as putting these questions in front of a human being, but it gives me an initial idea of how people might respond to these lines of questioning.
Now let me show you something even cooler. Watch this:
Please summarize the findings of the focus group transcript you just generated in a 1-page topline brief that draws out key insights. Address it to my client, Mr. Catson Dogwood.
Not bad for something that came from questions a computer generated and data it created! But the question I hope you’re asking here is, “if it can do this with fake data… what can it do with real data?” And the answer is pretty simple – if you were to copy in your focus group transcript a question at a time and then ask ChatGPT to summarize it into a one-page brief, it absolutely could… in mere seconds.
But let me show you something else it can do. Once again, I’m going to start by asking it to create something for me.
Pretend I just had a meeting with the marketing team for a company that makes specialty tires for United States automobiles. They are getting ready to launch a new product that can grip the road more tightly when there’s ice and prevent car accidents due to ice-related mishaps. The marketing team met for 30 minutes and talked about their goals for an upcoming survey with consumers to gauge interest in these tires. The team also wants to understand what brands consumers are aware of in this product category and what their decision process is. There are four people at the meeting: the director of sales, the digital marketing manager, the brand manager and the insights researcher.
Write a transcript of the meeting.
Can you believe it’s generating this meeting, complete with distinct characters and voices? I’m still amazed by this. But look. We have a transcript. Let’s use it to do a few useful things.
Let’s start by getting a meeting summary we can email to everyone, but it needs to be short:
Please write me some notes from this meeting, summarizing what we want to accomplish in one paragraph
Now, let’s see if ChatGPT can write a research purpose statement for us and identify some objectives!
Please use the transcript to write up a research purpose statement and also any associated research objectives that will shape the survey we intend to create
Oops! It wrote too many objectives. Let’s try to distill that down.
Let’s distill that down to 3-4 research objectives
Much better. I like to label my research objectives as well.
Please label each research objective in bold
OK, now for the part that is absolutely mind-blowing. I’m going to have ChatGPT write a survey for me based on these objectives.
Using the research purpose and the four objectives, please write a survey guide for a 5-minute online survey that will be administered to a national sample of adults aged 35-65 who own at least one vehicle and who live in areas where they are concerned about icy roads. Please also make sure to collect demographic information at the end of the survey on the participants’ gender, ZIP code, education, size of household and type of vehicle.
It actually took me longer to write the prompt than it did for ChatGPT to write that survey. And look! It’s a full survey, complete with an introduction, sections labeled by objective, appropriate scales for the questions asked and a length that is suited to a 5-minute window. The only thing I really see that it left out is a screener, which I can instruct it to add in:
Please regenerate the survey by including a screener for three criteria:
-Age (must be between 35 to 55; terminate if below 35 or over 55)
-Vehicle ownership (must own at least one automobile; terminate for no)
-Icy roads (must be concerned about icy roads; terminate if no)
Wow! It never fails to impress me how quickly and accurately ChatGPT can create things like this. And keep in mind, this is all from something we just made up a couple of minutes ago. If you were to give it actual meeting notes and actual objectives, can you imagine how much better it would be?
Here’s one more fun thing we can try for now: pilot testing our survey with fake respondents. So let’s try this input. I’m actually not even sure this is going to work.
Generate 10 responses to the above survey and give me an analysis of whether or not any questions need to be revised to collect better data
Holy smokes. It did exactly what I asked and even provided an assessment of the survey it wrote and told me how to improve it. Just amazing.
Are you excited? Are you scared? Is your mind reeling with the possibilities here?
Should I, as a marketing research and insights professional, be alarmed that ChatGPT can generate such a great survey so quickly?
And we are just scratching the surface. Just wait until our future videos, where I show you what ChatGPT4 can do.
With just what I’ve shown you today, you can probably already see the potential for using ChatGPT not just for fun, but to improve your capabilities. Just remember – if you update something to this program, we have no idea what OpenAI is keeping as future training data, so always make sure it’s disguised properly or that you have permission to upload it.
The next video in our series will be about using ChatGPT4 Plus and its plugins. If you thought what you saw today was incredible, you still haven’t seen anything yet. Future videos are going to focus on other AI programs like Stable Diffusion and Midjourney. Don’t miss them, and be sure to subscribe to this channel and hit the bell icon so you can be notified about future videos and you won’t miss out!
I’d love to hear what you think of this video in the comments, especially if you’re a marketing researcher or insights professional like me. And don’t be shy about sharing the cool things you’re doing with AI! See you next time!