Have you tried using ChatGPT, yet? If not, do not fret it. Perhaps you are a little like me. Given my own experience in the technology space and particularly, my work with putting AI to deliver business results, I am usually a bit wary of new technologies. As a junior in college, I read about the next big thing which included 3D TVs that offered ‘lifelike’ experiences right in our living rooms. I also gave Second Life –which was hyped-up as a quantum leap in social networking, a try. I also remember Google’s marketing push to mainstream Google Wave –a virtual platform communication and collaboration that would, among other things, replace emails. If you haven’t heard of 3D TVs or Second Life or Google Wave, the reason is evident: they were overhyped but turned out to be duds. (I will address the features and forces that make or break a new technology in a separate post.)
Therefore, on most days I refrain from investing time and effort to look up the “latest and best-est” technology being marketed out there. In the case of ChatGPT, however, I could not escape the everyday chatter –both virtual and real, around it. So about two months ago, when I found an opportunity I signed up to try ChatGPT (and you can too, here: http://chat.openai.com). I was not at all disappointed.
My first experience with an AI chatbot was in 2001. In a serendipitous turn of events, I happened to get my hands on a simple chatbot designed for high school children, to keep them company. A sort of simple babysitter or a virtual friend with which you can converse and it would dole out advise (even when you did not ask). But ChatGPT is anything but simple. It was different from all the chatbots I had experienced before –a lot more sophisticated along different dimensions of what one would normally label as intelligence.
Even when I know that I am just dealing a very complex interweave of artificial neurons and a loosely linked and compartmentalized transformer algorithms, the speed and quality of ChatGPT’s output to every day questions stands out.
In this post, I want to share with you the different things I tried with ChatGPT. I hope that after reading this post you will come up with your own ways to work with ChatGPT. I will save the usual caveats and disclaimers for later.
So, let’s get going!
1. Periodic Newsletters: A monthly or weekly newsletter used to be a very popular means to engage and attract audience. I asked ChatGPT to write a newsletter on the Greenblatt’s method of investing. Here is a cropped snapshot of what it generated.
2. Write Code for Simple Applications: by specifying the purpose, platform and language users can task ChatGPT to write a piece of code for simple applications. First, I asked it write a program in Java to generate the 1,000 prime numbers within a given range; but this seemed easy. In my second attempt, I asked it to write a code for a simple word counter application for the web (a web app) and this is what I got: What is impressive here is that ChatGPT presented a very structured answer which neatly organized the different code scripts for frontend and the backend of the web app. How great is that. (I’m no expert on Flask to test this, but I reckon this script should work directly as-is or with minor tweaks). I also discovered that a user had used ChatGPT to build a simple, but functional Twitter bot. Check it out here: https://tinyurl.com/2axusada
3. Design a Fitness Plan for You: Continuing to explore what else ChatGPT could do, I tried this. I hypothesized a scenario where a person is looking to shed weight only by reducing caloric intake. I specified the subject’s gender, height, weight and age and let the AI come out with some recommendations. Let us see what it came up with:
4. Write YouTube Scripts: Given that there are so many content creators out there, I was wondering if ChatGPT can automate some of the tasks that content creators carry out –such as writing a script. I asked ChatGPT to write a YouTube script for introducing decision trees to a bunch of college juniors (did not specify STEM / non-STEM to keep it simple) and this is what I got: A content script indeed comes with many caveats. The content creator / influencer has built a para-social relationship thanks largely to the unique connection between the influencer and her subscribers, so using a system generated script as-is is not a good idea. However, in terms boosting productivity this is a great place to start.
5. Accelerating Web Marketers and Content Planner Tasks: I asked myself if ChatGPT could help accelerate planning in the digital marketing space. I asked it to generate a content plan for my e-commerce portal selling pet toys for cats and dogs. I tried doing a lot more in this area of digital marketing with ChatGPT. I tasked it to come out with a general SEO plan, and asked it to narrow down on specific areas. Further, I used it to generate marketing catchphrases and copywriting content. I also used it to generate long form content. I must say that ChatGPT generated content was more than satisfactory for a general use case. It took about 20 to 30 minutes try these out. If starting from scratch, it would probably take about 12 to 20 hours of work. Needless to say, the AI generated content must be vetted, edited, adapted by the person using it. However, it probably saved at least few hours for the person doing it (something to think about: can a two person job be done by one + ChatGPT?).
6. Writing Topical Plan for Small e-Books: If you happen to peruse Amazon’s Kindle store, you will find hundreds, maybe even thousands, of kindle eBooks offered for anything between $0.99 and $4.99. Such books are mostly listed under the “Self-Help” section and are often authored by college seniors or freelance writers, and cover niche topics such as: a) Things to avoid during a family vacation with kids, or b) How to get the most out a fitness regimen, or c) Which college degree is best for you, and so on. I tried using ChatGPT to generate framework for such an eBook and this is what I got:
7. Writing a Play or a Movie Script: Can ChatGPT create a movie script. Surely not. Almost anyone with expertise in the area would say that AI agents like the ChatGPT can only repackage words and phrases in forms that manage to imply a commonly understood meaning in a relatively homogenous population. But if something seems new to a person, is it different from being new. To paraphrase Roger Penrose: if I were to type a character “A” on blank screen, delete it, and a few seconds later type “A” again on the same screen, is that the new A or the old one? Physicists taking the quantum line of reasoning would say there is no difference between the first A and the second one, while Newtonian physicists would say the second A is different from the first because of temporal stamping.
Without getting philosophical about it, let me share my experience with ChatGPT and about the creative side of things. I tried going a little deeper with ChatGPT and engaged it to write a movie script. The task was to write the plot for an adventure movie where a helpless prince is stranded on the moon and is rescued by a street-smart girl scientist and her 3 friends. It generated the plot and broke it up into five acts. It also generated character profiles of the Prince (“Max”) and the girl scientist (“Maya”) and her three friends. I could go into more details of my experience here, but during my research, I found that another user has already documented his experience with this. Check out Sung Kim’s experience documented here https://tinyurl.com/3h636236. (Now imagine creative writers combining the power of ChatGPT with DALL-E.)
Note: AI experts will argue, and correctly so, that the AI agent (ChatGPT) has not created something new from scratch, and that it has merely rearranged the text fed to it (i.e. it has been trained on trillions of words) in a form that gives you/us an impression of something new being created. Yes, we are very far from AI agents creating something truly new without first being trained on related data, but the question is –can a layperson know the difference between the two outputs? This is a question which will lead to many more questions than answers, and the potential answers to this will overlap the domains of philosophy and epistemology.
I will cover the last three things that I tried doing with ChatGPT (and you can too) in a separate post. In the next piece I will cover my experience with: 8) Gaming, 9) Trying to debug simple code, 10) Making WebMD obsolete. In that post I will also include a few things where ChatGPT fell far short of expectations. In short: while ChatGPT can do impressive things on the NLP frontier, it has major shortcomings in the quantitative side of things. Critical evaluation, logical assessment and analytical reasoning are not its strong suits. I will cover that in a separate post.
Prof. Abhijith S
Assistant Professor-Analytics & Data Science