Example 1 – “Find Bug in Code Screenshot”
In the following ChatGPT interaction, I tried pasting an image of a code snippet with a bug. Note that I didn’t paste the code, only the screenshot of the code as an image! I also gave it the prompt “Find the bug!“.
First, it does fix the bug, i.e., inserting the double equal symbol == in the comparison operator instead of the single assignment symbol =.
Yeah, the f-string explanation is clearly wrong so not a perfect hit in the first example. Still impressive and useful!
Example 2 – “Write HTML/CSS Website with Design Like Screenshot”
Next, I tried asking it to develop a website frontend based on nothing but a screenshot of the website. I used the Reddit frontend page at the time of writing to get started:
ChatGPT not only discovered that this is the Reddit homepage, it also started spitting out front-end code right away. This is insane value that you’d have paid $1,000 or so to a skilled developer only a year ago! 🤯
But does it work? Let’s copy the HTML and CSS to a file on my Desktop and open it:
Not bad, indeed. Very basic but I can further iterate with ChatGPT using the usual prompting tricks.
🧑💻 Recommended: Best 35 Helpful ChatGPT Prompts for Coders (2023)
Example 3 – Reverse Engineering Image Prompts
One of the highest-voted posts on Reddit today is the following image created by DALL-E. Can ChatGPT figure out the prompt that leads to this image? If yes, we can use it to reverse engineer prompting. Like prompt engineering goes prompt reverse engineering. 😉
ChatGPT spits out the following prompt:
"A neo-classical portrait of a Victorian gentleman and a celestial woman with historical tattoos, green eyes, and cosmic background"
In fact, it generates the following images based on the reverse engineered prompt — I’m impressed!
Example 4 – Create Code From Workflow Process
I created a simple workflow for processing events which is a very common computer science task:
The idea is to get an event, and assign it to a workflow balancer that sends the event to an event processor that, in turn, sends it back to the merger that outputs the processed events.
Can ChatGPT create a Python skeleton based on this workflow? Let’s see!
Yes! The Python code works and can be used to implement the workflow provided in the screenshot of my whiteboard drawings. This is amazing – ChatGPT could process and deeply understand all image experiments I came up with so far! 🚀🚀🚀
Example 5 – “Solve My Math Test”
Now let’s have a look at something with high utility for students and negative utility for teachers — if they don’t adapt! My daughter was blown away: 👇
I photographed a math test question and asked ChatGPT to solve it. It did with some inaccuracies (so be careful!):
And just like that, hundreds of years of “student culture” has evaporated. I expect the inaccuracies will be handled with the next model size – nobody said it was perfect yet.
I don’t recall how often I was stunned by technological disruption this year. It truly has been a mind-boggling year for humanity. Many of these tasks, like creating a front-end website or creating designs or code or texts, were priced at hundreds of dollars just a year ago. The value unlock for society already is significant – double-digit percentages in my estimation – but we’re just getting started. Five years from now, nothing will be the same.
If you want to stay on the right side of change, download our free coding cheat sheets and join our free email newsletter:
🔗 Recommended: Python OpenAI API Cheat Sheet (Free)
Also make sure to check out our learning path to become a prompt engineer from scratch:
🧑💻 Recommended: 10 Steps Learning Path: Becoming a Python Prompt Engineer 🚀 on the Finxter Academy