It might only be March, but without a doubt the cringe post of the year goes to -"you've probably heard about this new thing called ChatGPT", an embarrassingly common introduction line to posts on LinkedIn these days. We'll keep the recap as simple as possible as to not bore ourselves to death.
OpenAI, creators of AI text-to-image generation tool DALLE-2, dropped a new service called ChatGPT which takes text prompts written in plain English and submits them to a language model for processing. The algorithms then return what feels like a human response from the underlying data set which the model was trained on, and poof near magic unfolds.
In an arguably harrowing sense, the fanfare surrounding this newish technology is understandable, there's something incredibly unnerving about the whole thing. The depth and speed of the technology caught people off guard, it did and still does feel like it went from nothing to dystopian overnight. This resulted in a chorus of sorts. And then - almost in unison - the whole world lamented 'aaarghh Skynet', but alas they were wrong.
When using ChatGPT for the first time, you start from a place of healthy skepticism - you're entering text in a search bar like you do with Google, how much different can it be? Then bang, a mallet weighing 300 Billion words and 570GB of data twats you square in the chops with the force of a thousand exploding suns.
It's been quite some time since anyone in my generation has been bowled over by new technology, it's been the same for a while now - a new screen resolution here, a new processor or chip there - but mostly overall, it's been the same. So when ChatGPT came along everyone applied the same logic, the same thinking and the same way of interacting with technology as they had before.
Already this has lead to some sandbox style uses of the tool which are emblematic of any technology in its infancy. In search of the best use cases and admittedly by virtue of trying to avoid becoming noob myself, I decided to use ChatGPT only within the context of its initial purpose, as an assistant that communicates effectively, does the grunt work and ultimately makes humans more efficient.
Using ChatGPT Properly
Truth be told, there is no, singular 'right way' to use ChatGPT and there's nothing wrong with sandboxing or playing around; the idea here however is to show how and when it is best to use language models to improve efficiency and to learn something new, which are arguably the best times to use tools like ChatGPT in their current form.
Neutralising Imposter Syndrome
This is by far my favourite use case. We all suffer from Imposter Syndrome from time to time, in our daily lives we might proclaim some fact to be self-evident, only to second guess ourselves when someone disagrees. Other times we find ourselves lost when performing a work related task that we've done a thousand times before - sometimes you just need confirmation that your line of thinking is accurate and you haven't missed any obvious steps.
Don't Be Shy About It
As humans we have evolved to trim down our needs and requests so that they are more likely to be accepted by other humans, humans who have limits on the depth of their knowledge, their energy and their most precious resource - time. But remember language models aren't human, they only feel human, so you shouldn't impose the same limitations on what you ask for.
Now don't get me wrong, these are mostly god awful taglines, but I gave the model an obtuse ask and it still took less than 10 seconds to get pretty much what I wanted, bar some finesse in the execution.
Use It For Grunt Work
Another useful use case I found for ChatGPT was the loose end tasks that normally waste all of my spare time, such as writing meta descriptions for webpages which should be a 120-158 character summary of an entire topic or subject. When you find yourself with a rough, grizzly task to take care of - ChatGPT is built for that.
This whole interaction took less time than thinking about or writing a meta description for this article from scratch. Again it's not perfect, but it's enough for me to do the last 10% myself and merge the good and bad with some human judgement, and voila - I have what I need.
Developers Listen Up
On the face of it, using ChatGPT for your code seems like a dream, but beware - there are limitations. There have been a range of use cases where I've got some useful, small pieces of code and some great situations when I've used ChatGPT to get an explanation of where my code isn't playing ball - but also, when using ChatGPT to debug it doesn't have any understanding of the wider context of your application and thus, it gets things wrong quite often. It's not here to write your entire webapp code for you, or to wipe your knowledge gap shaped tears away - and no, it's not going to take your job either.
That said, when you get brain fog and reading the documentation just isn't cutting it, you can get a pretty clear response in layman terms for most issues. This saves a lot of time searching through the likes of StackOverflow for a solution to that unique bug that you just can't bash. This example was copied from a random repository on Github:
When ChatGPT Is The Noob
Watch Out, Bravado About
One thing I feel compelled to warn you about is ChatGPT's knack for providing information you've requested at any and all costs, including but not limited to providing reports with dates as references and then providing statistics which never existed as part of that report. This was a worrying observation because of the confidence ChatGPT had giving the initial answer - it would seem that ChatGPT has some imposter syndrome of it's own.
This is an issue that has been a worrying observation for competitors of ChatGPT too. Google recently experienced this first hand during a product demo of their 'Bard' language model, which gave a convincing yet entirely incorrect answer, leading to a $120B drop in Alphabet's market cap.
Outages, Outages Everywhere
Finally, if you've really got stuck into ChatGPT or you've incorporated it into your daily workflow, you've probably noticed that the service is sadly, painfully unreliable even outside of peak hours. If you're planning to use it at 4:55pm on a Friday afternoon then you'll probably realise you especially can't rely on it to be available when you need it the most.