How to use ChatGPT to automate the boring stuff no matter what your company does
After using ChatGPT for a month and writing several hundred lines of code with it. I have learned a few things:
The best analogy I can think of is Google in the late 90s. A personal story comes to mind:
Back in the late 90s my mom and sister were planning a trip, booking the flights for this trip required some knowledge from an airport in Washington, DC. The problem was that we lived in Chicago. A phone book for the Washington DC market was not readily available. "Just look it up online" I said.
Blank stares. "You can't just look things up like that online, we will need to make an expensive operator phone call" they told me.
I ran downstairs, connected to through dialup and searched this new tool called Yahoo! (powered by Google) returning with a phone number of what I thought was Reagan Airport but really was some random plumbing business in Northern Virginia.
When my sister called the number, the gentleman who answered had a good laugh over the mix up and then proceeded to provide the phone number for Reagan airport, clumsy but in the end she was able to solve the problem.
This story is one of the best analogies I can come up with to explain the state of programming and AI.
I've written HTML on and off since the late 90s, Learned python and various other web server stacks over the past few years as a hobby, but my only formal education in software programming are some undergrad and graduate level courses in C level embedded hardware programing. I've always been able to create basic scripts and simple programs but never really had the interest or need to create more advanced python programs or websites in HTML5 and Javascript.
Until I started playing with the chatbot.
I have long known that many processes at companies today can be automated with simple python scripts. In fact there is a whole book about it I read a number of years ago. https://automatetheboringstuff.com/ The issue with implementation was never the invention of a tool, Python is the tool for process and HTML and Javascript are the tools for dashboards to display the output. (along with many others)
The issue was the time it took to work up an idea, find the right python libraries and write the program.
领英推荐
This could take minutes sometimes or hours/days. You could think you had the perfect solution for automating how to read a PDF into a pandas dataframe and sort the information to build a quote or enter a sales order, but maybe some small part would go wrong or a deprecated library sending you back to a long series of datasheets and documentation (or Google).
Of course, there are plenty of software engineers who offer services to speed this process up, however this then takes time to plan out and create the solution without code, then pay a developer to write the code. It can also add to the costs and might make the idea not worth automating.
Enter AI
Now, with the widespread availability of a computer that writes code, the beginning of every sci-fi movie since the late 80s has come to life. Whoops, I mean, the enablement of testing and quickly deploying ideas is made possible.
In one afternoon, I was able to develop HTML and javascript code to display a youtube transcript that highlights as it plays and will allow the user to play the video when the corresponding part of the transcript is clicked. I paired this with a python script to pull a youtube video transcript, parse it into 5 second segments and output the html page that contains the completed project. It took a handful of iterations, and a number of AI hallucinations before complete but I was able to get a hacked together project finished. Here is a sample of the output, unpolished, but working:
The python program looked like this (I have since modified it a little)
While just a simple example, the ability to write HTML, Javascript and Python by machine open up a lot of doors for automation, reporting and data analysis. Stuff that is done today at high expense.
This is only the beginning and of course it is a sample free chatbot (there is also a paid version of GPT-3) that maybe can't be relied on indefinitely. One thing that is for certain though, is programming will change significantly shuffling the deck of work and ideas between product management, operations sales and engineering. (Imagine product managers that can test and get basic concepts of ideas before tossing them out to the engineering team to do so, or Sales teams that can generate their own reports from massive MRP datasets)
There are thousands of python libraries for every task you can imagine from Oracle databases to Youtube, What manual process are you thinking about automating but haven't taken the leap with yet?
Insight-led recruitment. UK key hires when your context matters. 40,000 hours in recruitment - the more I learn, the less I know.
1 年The current iteration certainly seems a good extension of a skilled operator, rather than any form of replacement. Learning how to get the best out of it is a good skill. Can't wait to see what will have developed, this time next year.